Rebuild Space as static no-GAT explorer (Ablation 2)

Replace the broken Gradio app (Jinja2 unhashable-dict crash) with a static
front-end for the current 'No Drug Graphs' model. Data is served cross-origin
from the project's GitHub Pages so the Space and Pages share one source of truth.

.gitattributes

@@ -1,35 +1,2 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.html linguist-language=HTML
+* text=auto eol=lf

.gitignore

@@ -1,68 +0,0 @@
-# Ignore everything by default
-
-
-# But not these files:
-!.gitignore
-
-# Allow CSV files
-!*.csv
-
-# Allow Jupyter notebooks
-!*.ipynb
-
-# Allow Python files
-!*.py
-
-# Allow directories (so we can traverse into them)
-!*/
-
-# Ignore common large files even if they match above patterns
-*.pkl
-*.npy
-*.db
-*.xml
-*.h5
-*.hdf5
-
-# Ignore model checkpoints and weights
-*.pth
-*.pt
-*.ckpt
-*.safetensors
-
-# Ignore image outputs
-*.png
-*.jpg
-*.jpeg
-*.gif
-*.svg
-
-# Ignore compressed files
-*.zip
-*.tar
-*.gz
-*.rar
-
-# Ignore Python cache
-__pycache__/
-*.pyc
-*.pyo
-*.pyd
-.Python
-
-# Ignore Jupyter checkpoints
-.ipynb_checkpoints/
-
-# Ignore environment files
-.env
-.venv
-env/
-venv/
-ENV/
-
-# Ignore IDE files
-.vscode/
-.idea/
-*.swp
-*.swo
-*~

APP_GUIDE.md

@@ -1,277 +0,0 @@
-# 🚀 Gradio App Quick Start Guide
-
-## What I Created
-
-I've built a **fully interactive Gradio web app** for your Pharmacology Knowledge Graph that includes:
-
-✅ **Drug search functionality** - Search by name or ChEMBL ID  
-✅ **Known interactions display** - View verified drug-target and drug-effect relationships  
-✅ **AI predictions** - See novel targets and therapeutic uses predicted by your model  
-✅ **Interactive network visualization** - Beautiful Plotly graph showing drug mechanisms  
-✅ **Ready for Hugging Face deployment** - One-click hosting on HF Spaces  
-
----
-
-## 📁 Files Created
-
-| File | Purpose |
-|------|---------|
-| `app.py` | Main Gradio application (500+ lines) |
-| `requirements_app.txt` | Python dependencies for the app |
-| `run_app.sh` | Local testing script |
-| `DEPLOYMENT.md` | Step-by-step HF Spaces deployment guide |
-| `SPACE_README.md` | README for your Hugging Face Space |
-
----
-
-## 🎮 How to Use It
-
-### Option 1: Test Locally (Recommended First)
-
-```bash
-# Run the app on your machine
-./run_app.sh
-
-# Or manually:
-python app.py
-```
-
-Then open: **http://localhost:7860**
-
-### Option 2: Deploy to Hugging Face Spaces
-
-Follow the guide in `DEPLOYMENT.md`. Summary:
-
-1. **Create Space** on huggingface.co/spaces
-2. **Upload files:**
-   - `app.py`
-   - `requirements_app.txt` → rename to `requirements.txt`
-   - All CSV files (drug_nodes, protein_nodes, etc.)
-   - All NPY files (graph_embeddings, node_to_idx)
-3. **Wait 2-3 minutes** for build
-4. **Share your link!** `https://huggingface.co/spaces/YOUR_USERNAME/pharmacology-knowledge-graph`
-
----
-
-## 🎨 App Features
-
-### 1. Drug Search
-- Type any drug name (e.g., "Morphine", "Aspirin")
-- Or use ChEMBL ID (e.g., "CHEMBL70")
-- Get auto-suggestions as you type
-
-### 2. Information Panels
-
-**Left Side:**
-- Drug basic info (name, ChEMBL ID, SMILES)
-- Known protein targets with binding affinity (pChEMBL)
-- Known clinical effects (indications)
-
-**Right Side:**
-- **Predicted novel targets** (drug repurposing opportunities)
-- **Predicted novel effects** (new therapeutic uses)
-- Confidence scores and similarity metrics
-
-### 3. Interactive Network Graph
-
-Shows your drug as the **center node** with:
-- 🔵 **Blue nodes** = Known protein targets (solid edges)
-- 🟢 **Green nodes** = Predicted targets (dashed edges)
-- 🟡 **Yellow nodes** = Known effects (solid edges)
-- 🟨 **Light yellow nodes** = Predicted effects (dashed edges)
-
-**Controls:**
-- Toggle known/predicted interactions on/off
-- Adjust number of nodes displayed (10-50)
-- Hover over nodes for details
-- Zoom/pan the graph
-
----
-
-## 📊 Example Queries to Try
-
-| Drug | What You'll See |
-|------|----------------|
-| **Morphine** | Opioid receptors (μ, δ, κ), analgesic effects, similar drugs like Fentanyl |
-| **Aspirin** | COX-1/COX-2 inhibition, anti-inflammatory effects, cardiovascular uses |
-| **Talazoparib** | PARP1/2 targets + **predicted** PARP3/4 homologs (novel finding!) |
-| **Metformin** | AMPK activation, diabetes treatment, **predicted** anti-aging effects |
-| **Rivaroxaban** | Known anticoagulation + **predicted** myocardial infarction prevention |
-
----
-
-## 🏗️ Technical Architecture
-
-```
-User Input (Drug Name)
-    ↓
-[Search Function] → Find drug in database
-    ↓
-[Embedding Lookup] → Get drug's 128-dim vector
-    ↓
-[Similarity Computation] → Cosine similarity to all proteins/effects
-    ↓
-[Ranking & Filtering] → Top-K predictions, remove known interactions
-    ↓
-[Network Builder] → Create interactive Plotly graph
-    ↓
-Display Results
-```
-
-**Key Components:**
-
-1. **Data Loading** (startup):
-   - CSV files → Pandas DataFrames
-   - NPY files → NumPy arrays (embeddings)
-   - Cached in memory for fast access
-
-2. **Search Engine**:
-   - Fuzzy matching on drug names
-   - ChEMBL ID exact matching
-   - Returns top 20 matches
-
-3. **Prediction Engine**:
-   - Cosine similarity: `cos(drug_emb, target_emb)`
-   - Threshold filtering: High (>0.5), Medium (>0.45), Low
-   - Excludes known interactions
-
-4. **Visualization**:
-   - NetworkX for graph layout (spring layout)
-   - Plotly for interactive rendering
-   - Color-coded by node type and known/predicted status
-
----
-
-## 🎯 Deployment Checklist
-
-### Pre-Deployment (Do This First)
-
-- [ ] Test app locally: `./run_app.sh`
-- [ ] Try 5-10 different drugs
-- [ ] Check network visualization works
-- [ ] Verify predictions make sense
-
-### Hugging Face Setup
-
-- [ ] Create HF account (if needed)
-- [ ] Create new Space
-- [ ] Choose "Gradio" SDK
-- [ ] Select CPU basic (free tier)
-
-### File Upload
-
-- [ ] Upload `app.py`
-- [ ] Rename `requirements_app.txt` → `requirements.txt` and upload
-- [ ] Upload all CSV files:
-  - [ ] `drug_nodes.csv`
-  - [ ] `protein_nodes_with_embeddings.csv`
-  - [ ] `drug_effects.csv`
-  - [ ] `drugs_interactions.csv`
-- [ ] Upload NPY files:
-  - [ ] `graph_embeddings.npy`
-  - [ ] `node_to_idx.npy`
-- [ ] Use `SPACE_README.md` content for Space README
-
-### Post-Deployment
-
-- [ ] Wait for build (check logs)
-- [ ] Test live app
-- [ ] Share link on social media
-- [ ] Add to competition submission
-
----
-
-## 💡 Pro Tips
-
-### For Competitions/Demos
-
-1. **Prepare example queries** - Have 3-5 drugs ready to show during demo
-2. **Highlight novel predictions** - Point out predicted interactions not in training data
-3. **Show validation** - Mention the ~90% precision rate
-4. **Explain the network** - Walk through how to interpret the visualization
-
-### For Presentations
-
-**30-second pitch:**
-> "This app lets you explore how drugs work using AI. Type any drug name, and you'll see known targets plus AI-predicted new uses. The network shows everything visually. We trained it on 800 drugs and validated 90% accuracy on predictions."
-
-**Key talking points:**
-- Novel approach: Learns from drug structure + protein sequence
-- Real data: 800 FDA drugs, 200 proteins, 400 effects
-- Practical use: Drug repurposing, target discovery, safety prediction
-- Validated: 90% of top predictions are biologically coherent
-
-### For Further Development
-
-Want to enhance it? Easy additions:
-
-```python
-# 1. Add SMILES input for novel molecules
-def predict_from_smiles(smiles_string):
-    # Use your Ridge regression model
-    # Return predictions
-    
-# 2. Export results to CSV
-def export_predictions(drug_name):
-    # Generate downloadable file
-    return gr.File(...)
-
-# 3. Batch analysis
-def analyze_multiple_drugs(drug_list):
-    # Loop and aggregate results
-```
-
----
-
-## 🐛 Troubleshooting
-
-### "Module not found" error
-```bash
-# Install missing package
-uv pip install <package_name>
-```
-
-### "File not found" error
-```bash
-# Check all CSV/NPY files are in same directory as app.py
-ls *.csv *.npy
-```
-
-### Slow performance
-- Predictions are cached after first run
-- Consider pre-computing all predictions (add CSV files)
-- Hugging Face Spaces: upgrade to CPU upgrade ($0.03/hr)
-
-### Network visualization not showing
-- Check Plotly is installed: `uv pip install plotly`
-- Try reducing max_nodes in settings
-
----
-
-## 📈 Next Steps
-
-1. **Test locally** → Run `./run_app.sh` and try it out
-2. **Deploy to HF** → Follow `DEPLOYMENT.md` guide
-3. **Share** → Get the public URL and share widely
-4. **Iterate** → Gather feedback, add features
-
----
-
-## 🎉 You're Ready!
-
-You now have a **professional, interactive web app** that:
-- ✅ Showcases your research
-- ✅ Is easy to use (no code required for users)
-- ✅ Deploys with one click
-- ✅ Looks impressive in competitions/demos
-
-**Questions?** Check:
-- `DEPLOYMENT.md` for HF Spaces setup
-- `app.py` comments for code details
-- GitHub issues for community support
-
----
-
-**Good luck with your competition! 🚀**
-
-*P.S. - Once deployed, share the link! People love interactive ML demos.*

DEPLOYMENT.md

@@ -1,221 +0,0 @@
-# Hugging Face Space Deployment Guide
-
-## 🚀 Deploy to Hugging Face Spaces
-
-### Step 1: Create a New Space
-
-1. Go to [Hugging Face Spaces](https://huggingface.co/spaces)
-2. Click "Create new Space"
-3. Choose:
-   - **Space name:** `pharmacology-knowledge-graph`
-   - **License:** MIT
-   - **Space SDK:** Gradio
-   - **Space hardware:** CPU basic (free tier works fine)
-
-### Step 2: Upload Required Files
-
-Upload these files to your Space:
-
-**Required files:**
-```
-app.py                                  # Main Gradio app
-requirements_app.txt                    # Python dependencies (rename to requirements.txt)
-drug_nodes.csv                          # Drug metadata
-protein_nodes_with_embeddings.csv       # Protein data
-drug_effects.csv                        # Clinical effects
-drugs_interactions.csv                  # Known drug-protein interactions
-graph_embeddings.npy                    # Trained embeddings
-node_to_idx.npy                         # Node index mapping
-top_50_predicted_drug_protein.csv       # Pre-computed predictions (optional)
-top_50_predicted_drug_effects.csv       # Pre-computed predictions (optional)
-```
-
-**File structure on Hugging Face:**
-```
-your-space/
-├── app.py
-├── requirements.txt                    # (rename requirements_app.txt)
-├── README.md                           # Space description
-├── drug_nodes.csv
-├── protein_nodes_with_embeddings.csv
-├── drug_effects.csv
-├── drugs_interactions.csv
-├── graph_embeddings.npy
-├── node_to_idx.npy
-├── top_50_predicted_drug_protein.csv
-└── top_50_predicted_drug_effects.csv
-```
-
-### Step 3: Create Space README.md
-
-Create a `README.md` in your Space with this content:
-
-```markdown
----
-title: Pharmacology Knowledge Graph Explorer
-emoji: 💊
-colorFrom: blue
-colorTo: green
-sdk: gradio
-sdk_version: 4.44.0
-app_file: app.py
-pinned: false
-license: mit
----
-
-# 💊 Pharmacology Knowledge Graph Explorer
-
-Explore drug-target-effect relationships using AI-powered predictions from a TransE knowledge graph model.
-
-## Features
-
-- 🔍 Search 800+ FDA-approved drugs
-- 🎯 View known and predicted protein targets
-- 💉 Discover potential therapeutic uses (drug repurposing)
-- 🕸️ Interactive network visualization
-
-## How It Works
-
-This app uses a **TransE knowledge graph embedding model** trained on:
-- 800+ FDA-approved drugs (from ChEMBL)
-- 200+ human protein targets (with ESM-2 embeddings)
-- 400+ clinical effects and indications
-
-**Prediction method:** Cosine similarity in learned embedding space  
-**Model performance:** ~90% precision on top-50 predictions
-
-## Example Queries
-
-Try searching for:
-- **Morphine** - See opioid receptor targets and analgesic effects
-- **Aspirin** - COX inhibition and cardiovascular effects  
-- **Metformin** - Diabetes and potential repurposing candidates
-
-## Citation
-
-```bibtex
-@software{pharmacology_graph_2025,
-  author = {Joe VonDahab},
-  title = {Pharmacology Knowledge Graph: Drug-Target-Effect Prediction},
-  year = {2025},
-  url = {https://github.com/JoeVonDahab/pharmacology-graph}
-}
-```
-
-## Disclaimer
-
-This is a research tool for exploratory analysis only. Predictions should be validated experimentally. Not for clinical use.
-```
-
-### Step 4: File Preparation Commands
-
-Run these commands in your project directory:
-
-```bash
-# 1. Copy app requirements (rename for HF)
-cp requirements_app.txt requirements.txt
-
-# 2. Verify all data files exist
-ls -lh *.csv *.npy
-
-# Expected files:
-# - drug_nodes.csv (~200KB)
-# - protein_nodes_with_embeddings.csv (~500KB)
-# - drug_effects.csv (~300KB)
-# - drugs_interactions.csv (~2MB)
-# - graph_embeddings.npy (~700KB)
-# - node_to_idx.npy (~50KB)
-```
-
-### Step 5: Upload to Hugging Face
-
-**Option A: Web Upload**
-
-1. Go to your Space's Files tab
-2. Click "Add file" → "Upload files"
-3. Drag and drop all files
-4. Commit changes
-
-**Option B: Git Upload**
-
-```bash
-# Clone your space
-git clone https://huggingface.co/spaces/YOUR_USERNAME/pharmacology-knowledge-graph
-cd pharmacology-knowledge-graph
-
-# Copy files
-cp /path/to/pharmacology-graph/app.py .
-cp /path/to/pharmacology-graph/requirements_app.txt requirements.txt
-cp /path/to/pharmacology-graph/*.csv .
-cp /path/to/pharmacology-graph/*.npy .
-
-# Commit and push
-git add .
-git commit -m "Initial app deployment"
-git push
-```
-
-### Step 6: Monitor Deployment
-
-1. The Space will automatically build (takes ~2-3 minutes)
-2. Check the build logs for errors
-3. Once running, test the app with example drugs
-
-### Step 7: Optional Enhancements
-
-**Upgrade to GPU (for faster predictions):**
-- Settings → Hardware → Upgrade to T4 small ($0.60/hour)
-- Useful if computing predictions on-the-fly for new molecules
-
-**Enable persistence:**
-- Add a `cache/` directory for storing results
-- Set `GRADIO_CACHE_EXAMPLES=True` in Space settings
-
-**Add analytics:**
-```python
-# In app.py, add Hugging Face analytics
-import os
-from huggingface_hub import HfApi
-
-# Track usage
-api = HfApi()
-```
-
-### Troubleshooting
-
-**"File not found" errors:**
-- Make sure all CSV and NPY files are in the root directory
-- Check file names match exactly (case-sensitive)
-
-**Memory errors:**
-- If embeddings are too large, upgrade to "CPU upgrade" ($0.03/hour)
-- Or pre-compute all predictions and use lookup tables
-
-**Slow loading:**
-- Add caching: `@st.cache_data` (Streamlit) or `gr.State()` (Gradio)
-- Pre-load data in global scope (already done in `app.py`)
-
-### Cost Estimate
-
-**Free tier (CPU basic):**
-- ✅ Sufficient for this app
-- 2 vCPU, 16GB RAM
-- Always-on
-
-**Paid tier (if needed):**
-- CPU upgrade: $0.03/hour (~$22/month)
-- T4 GPU: $0.60/hour (only if doing real-time SMILES predictions)
-
----
-
-## 🎉 You're Done!
-
-Your app should now be live at:
-`https://huggingface.co/spaces/YOUR_USERNAME/pharmacology-knowledge-graph`
-
-Share it with:
-- Colleagues and collaborators
-- On Twitter/LinkedIn with #DrugDiscovery #AI
-- In your competition submission
-
-Good luck! 🚀

FILES.md

@@ -1,214 +0,0 @@
-# File Manifest - Pharmacology Graph
-
-Quick reference for what each file does.
-
----
-
-## 🚀 **ESSENTIAL FILES** (Keep These)
-
-### Application Files
-- **`app.py`** - Main Gradio web application (604 lines)
-- **`requirements_app.txt`** - Python dependencies for the app
-- **`start_app.sh`** - Launch script (recommended)
-- **`test_app.py`** - Diagnostic test for data loading
-
-### Data Files (Required)
-- **`drug_nodes.csv`** - 3,127 approved drugs with SMILES structures
-- **`protein_nodes_with_embeddings.csv`** - 1,156 protein targets
-- **`drug_effects.csv`** - 8,312 clinical effects/indications
-- **`drugs_interactions.csv`** - 11,703 known drug-protein interactions
-- **`graph_embeddings.npy`** - Trained 128-dim TransE embeddings (5,201 nodes)
-- **`node_to_idx.npy`** - Node ID to embedding index mapping
-- **`top_50_predicted_drug_protein.csv`** - Top 50 predicted drug-target pairs
-- **`top_50_predicted_drug_effects.csv`** - Top 50 predicted drug-effect pairs
-
-### Documentation
-- **`README.md`** - Project overview, results, and citations
-- **`SETUP.md`** - Complete setup guide (START HERE!)
-- **`DEPLOYMENT.md`** - Guide for deploying to Hugging Face Spaces
-- **`APP_GUIDE.md`** - User guide for the web application
-
----
-
-## 🔬 **DEVELOPMENT FILES** (Optional)
-
-### Research Notebook
-- **`code.ipynb`** - Main Jupyter notebook with full pipeline:
-  - Data extraction from ChEMBL
-  - ESM-2 protein embedding generation (~30 min GPU)
-  - TransE graph model training (100 epochs)
-  - Link prediction and validation
-  
-**Note**: Only needed if you want to retrain the model or explore the data processing pipeline.
-
-### Full Dependencies
-- **`requirements.txt`** - Complete dependencies (includes transformers, torch, rdkit)
-  - **Use this if**: You want to run `code.ipynb`
-  - **Skip if**: You just want to run the app (use `requirements_app.txt` instead)
-
-### ChEMBL Database
-- **`chembl_36/chembl_36_sqlite/chembl_36.db`** - Full ChEMBL 36 database (4.2GB)
-  - **Needed for**: Running `code.ipynb` to extract fresh data
-  - **Skip if**: You're using the pre-extracted CSV files
-
----
-
-## 🗑️ **REMOVABLE FILES** (Can Delete)
-
-### Duplicates/Old Versions
-- `run_app.sh` - Old launch script (use `start_app.sh` instead)
-- `APP_SUMMARY.md` - Duplicate of `APP_GUIDE.md`
-- `CHECKLIST.md` - Deployment checklist (one-time use)
-- `BUGFIX_SUMMARY.md` - Development notes about version fixes
-
-### Testing Scripts
-- `test_search.py` - Manual testing guide (not automated)
-
-### Intermediate Files (Can Regenerate)
-- `main.py` - Old/experimental code
-- `.python-version` - Python version pin (use pyproject.toml instead)
-- `*.png` - Visualization outputs (can regenerate from notebook)
-- `*.pkl` - Pickle serializations (can regenerate from CSV)
-- `drug_mechanism_filtered.csv` - Intermediate processing file
-- `drug_warnings.csv` - Intermediate processing file
-- `edges_drug_protein.csv` - Intermediate processing file
-- `proteins_for_embedding.csv` - Intermediate processing file
-- `top_50_predicted_drug_targets.csv` - Duplicate of `top_50_predicted_drug_protein.csv`
-
-**Run cleanup script**: `./cleanup.sh` to remove these automatically.
-
----
-
-## 📏 **FILE SIZES**
-
-### App Files (~10 MB total)
-```
-drug_nodes.csv                          700 KB
-protein_nodes_with_embeddings.csv       400 KB
-drug_effects.csv                        800 KB
-drugs_interactions.csv                  1.5 MB
-graph_embeddings.npy                    5 MB
-node_to_idx.npy                         50 KB
-top_50_predicted_*.csv                  10 KB each
-app.py                                  50 KB
-```
-
-### Development Files (4.2 GB)
-```
-chembl_36/chembl_36_sqlite/chembl_36.db  4.2 GB
-code.ipynb                               500 KB
-```
-
----
-
-## 🎯 **WHAT YOU NEED FOR...**
-
-### Running the App Only
-```
-✓ app.py
-✓ requirements_app.txt
-✓ start_app.sh (or test_app.py)
-✓ All CSV files
-✓ graph_embeddings.npy
-✓ node_to_idx.npy
-✓ README.md, SETUP.md (for reference)
-```
-**Total**: ~10 MB
-
-### Retraining the Model
-```
-✓ Everything from "Running the App"
-✓ code.ipynb
-✓ requirements.txt
-✓ chembl_36/ directory (ChEMBL database)
-```
-**Total**: ~4.2 GB
-
-### Deploying to Hugging Face Spaces
-```
-✓ app.py
-✓ All CSV files
-✓ graph_embeddings.npy
-✓ node_to_idx.npy
-✓ requirements_app.txt → rename to requirements.txt
-✓ SPACE_README.md → use as README.md
-```
-**Total**: ~10 MB (HF Spaces free tier supports up to 50GB)
-
----
-
-## 🧹 **CLEANUP COMMANDS**
-
-### Remove all unnecessary files
-```bash
-./cleanup.sh
-```
-
-### Manual cleanup
-```bash
-# Remove duplicate documentation
-rm -f BUGFIX_SUMMARY.md APP_SUMMARY.md CHECKLIST.md
-
-# Remove old scripts
-rm -f run_app.sh test_search.py main.py
-
-# Remove generated outputs
-rm -f *.png *.pkl
-
-# Remove intermediate CSV files
-rm -f drug_mechanism_filtered.csv drug_warnings.csv edges_drug_protein.csv proteins_for_embedding.csv
-```
-
-### Remove development files (if you don't need to retrain)
-```bash
-# Remove notebook and database (WARNING: Large files!)
-rm -rf code.ipynb chembl_36/ requirements.txt
-```
-
----
-
-## 📦 **WHAT'S IN GIT**
-
-The `.gitignore` is configured to:
-
-**✅ Track:**
-- All Python files (`*.py`)
-- All CSV files (`*.csv`)
-- All Jupyter notebooks (`*.ipynb`)
-- Documentation (`*.md`)
-- Config files (`requirements*.txt`, `pyproject.toml`)
-
-**❌ Ignore:**
-- Large binary files (`*.npy`, `*.pkl`, `*.db`)
-- Images (`*.png`, `*.jpg`)
-- Python cache (`__pycache__/`, `*.pyc`)
-- Model checkpoints (`*.pth`, `*.pt`)
-- Compressed files (`*.zip`, `*.tar.gz`)
-
-**Why NPY files aren't tracked**: They're 5MB+ and can be regenerated from the notebook. Download them separately or run the notebook to create them.
-
----
-
-## 🔄 **REGENERATING FILES**
-
-If you're missing `.npy` files or want to retrain:
-
-```bash
-# Install full dependencies
-pip install -r requirements.txt
-
-# Open notebook
-jupyter notebook code.ipynb
-
-# Run all cells (takes ~40 minutes on GPU)
-# This will regenerate:
-# - All CSV files
-# - graph_embeddings.npy
-# - node_to_idx.npy
-# - Prediction CSVs
-```
-
----
-
-**Last Updated**: October 10, 2025  
-**Total Project Size**: 4.2 GB (with database) or 10 MB (app only)

README.md

@@ -1,415 +1,29 @@
 ---
-title: Pharmacology Knowledge Graph
+title: Pharmacology Graph Explorer
 emoji: 💊
 colorFrom: blue
 colorTo: purple
-sdk: gradio
-sdk_version: 4.44.1
-app_file: app.py
+sdk: static
+app_file: index.html
 pinned: false
 license: mit
-python_version: "3.10"
 ---
 
-# Pharmacology Knowledge Graph: Drug-Target-Effect Prediction
+# Pharmacology Graph Explorer
 
-A novel machine learning system that learns **pharmacology-aligned embeddings** by integrating chemical structures (SMILES), protein sequences (ESM-2), and clinical outcomes into a unified knowledge graph. The model predicts novel drug-target interactions and therapeutic indications through contrastive graph learning.
+Interactive explorer for the **Ablation 2 "No Drug Graphs"** model (~3.29M params) from
+*pharmacology-graph* — learnable drug embeddings + ESM-2 protein representations + heterogeneous
+GraphSAGE message passing. Removing the GAT drug-structure encoder *improved* drug–protein
+prediction (PR-AUC 0.5631 → 0.5785) while cutting memory 5.30 GB → 353 MB.
 
-**🚀 [Try the Interactive Demo](https://huggingface.co/spaces/JoeVonDahab/pharmacology-graph)** | **📖 [Full Setup Guide](SETUP.md)** | **🎓 [Research Notebook](code.ipynb)**
+Search a **drug** to see predicted targets & indications, or an **indication / protein** to see the
+drugs most likely linked to it — each flagged *known* or *novel*.
 
----
-
-## ⚡ Quick Start
-
-```bash
-# 1. Clone the repository
-git clone https://github.com/JoeVonDahab/pharmacology-graph.git
-cd pharmacology-graph
-
-# 2. Install dependencies
-pip install -r requirements_app.txt
-
-# 3. Run the interactive app
-./start_app.sh
-# Open http://localhost:7860 in your browser
-```
-
-**Requirements**: Python 3.9+, 8GB RAM, 10MB disk space (app only)
-
-For complete setup instructions and troubleshooting, see **[SETUP.md](SETUP.md)**.
-
----
-
-## 🎯 Project Overview
-
-This project builds an end-to-end pipeline that:
-
-1. **Extracts pharmacological data** from ChEMBL database (approved drugs, protein targets, clinical effects)
-2. **Generates molecular representations** using:
-   - ESM-2 protein language models (2560-dim embeddings)
-   - SMILES-based molecular fingerprints
-3. **Trains a TransE graph embedding model** to align drugs, proteins, and clinical effects in a shared latent space
-4. **Predicts novel interactions** using GPU-accelerated similarity search
-
-### Key Innovation: Pharmacology-Aligned Embeddings
-
-Unlike traditional chemical similarity (fingerprints) or docking approaches, our model learns embeddings where **drugs cluster by functional mechanism** rather than just structural similarity. For example:
-
-- **Opioid analgesics** (Morphine, l) cluster together despite different structures
-- **Antivirals** (Acyclovir, Famciclovir) group by shared viral DNA polymerase targets
-- **Bronchodilators** (Tiotropium, Ipratropium) align based on muscarinic receptor activity
-
----
-
-## 🏗️ Architecture
-
-```
-ChEMBL Database
-    ↓
-[Data Extraction]
-    ├── Approved Drugs (SMILES)
-    ├── Protein Targets (sequences)
-    └── Clinical Effects (indications)
-    ↓
-[Embedding Generation]
-    ├── ESM-2 (facebook/esm2_t36_3B_UR50D) → Protein embeddings
-    └── Morgan Fingerprints → Drug structure features
-    ↓
-[Knowledge Graph Construction]
-    ├── Nodes: Drugs, Proteins, Effects
-    └── Edges: binds_to, treats
-    ↓
-[TransE Training]
-    └── 128-dim unified embedding space
-    ↓
-[Link Prediction]
-    ├── Drug → Protein (novel targets)
-    └── Drug → Effect (repurposing)
-```
-
----
-
-## 📊 Results
-
-### Validation: Drug Clustering
-
-The model successfully groups drugs by pharmacological class:
-
-| Anchor Drug | Top Similar Drugs | Pharmacological Class |
-|-------------|-------------------|----------------------|
-| **Morphine** | l, Oxycodone, Hydromorphone | Opioid analgesics (μ-receptor) |
-| **Acyclovir** | Famciclovir, Penciclovir, Valacyclovir | Nucleoside antiviral (DNA polymerase) |
-| **Tiotropium** | Glycopyrronium, Ipratropium, Aclidinium | Anticholinergic bronchodilators (M3-receptor) |
-| **Miconazole** | Posaconazole, Ketoconazole | Azole antifungals (CYP51A1) |
-
-**Mean cosine similarity:**
-- Within therapeutic class: **0.72 ± 0.08**
-- Between classes: **0.31 ± 0.12**
-
-### Top Predicted Drug-Target Interactions
-
-From **2.8M novel predictions**, top examples with biological validation:
-
-| Drug | Predicted Target | Similarity | Biological Rationale |
-|------|-----------------|------------|---------------------|
-| **Talazoparib** | Protein mono-ADP-ribosyltransferase (PARP3/4) | 0.60 | ✅ Known PARP1/2 inhibitor; homologous family members |
-| **Dasatinib** | Blk tyrosine kinase | 0.51 | ✅ Broad-spectrum Src-family kinase inhibitor |
-| **Imipramine** | α1D adrenergic receptor | 0.52 | ✅ Tricyclic with known adrenergic off-targets |
-| **Zonisamide** | Carbonic anhydrase 14 | 0.51 | ✅ Known CA inhibitory activity |
-| **Pipamazine** | Muscarinic M4 receptor | 0.54 | ✅ Phenothiazine with anticholinergic effects |
-
-### Top Predicted Drug-Effect (Repurposing Candidates)
-
-| Drug | Predicted Indication | Similarity | Clinical Plausibility |
-|------|---------------------|------------|---------------------|
-| **Rivaroxaban** | Myocardial infarction (secondary prevention) | 0.54 | ✅ Anticoagulant; approved for post-MI use |
-| **Sertraline** | Panic disorder | 0.53 | ✅ FDA-approved indication |
-| **Rosuvastatin** | Dyslipidemias | 0.52 | ✅ Primary statin indication |
-| **Ozanimod** | Crohn's disease | 0.49 | ⚠️ Plausible (approved for ulcerative colitis) |
-| **Clopidogrel** | Pulmonary embolism | 0.49 | ⚠️ Antiplatelet; not first-line but mechanistically coherent |
-
-**Precision metrics:**
-- Top-50 predictions: **~90% pharmacologically coherent**
-- Mean similarity (novel predictions): **0.48 ± 0.05**
-- Baseline (random pairing): **0.23 ± 0.11**
-
----
-
-## 🚀 Quick Start
-
-### Requirements
-
-```bash
-# Install dependencies with UV
-uv pip install -r requirements.txt
-
-# Core dependencies:
-- pandas
-- numpy
-- torch
-- transformers (ESM-2)
-- rdkit
-- networkx
-- scikit-learn
-- tqdm
-- matplotlib
-```
-
-### Data
-
-Download ChEMBL 36 SQLite database:
-```bash
-# Place in: chembl_36/chembl_36_sqlite/chembl_36.db
-# Size: ~4.2 GB
-```
-
-### Run the Pipeline
-
-```bash
-# Open the Jupyter notebook
-jupyter notebook code.ipynb
-
-# Or run as Python script (convert cells first)
-jupyter nbconvert --to script code.ipynb
-python code.py
-```
-
-**Pipeline stages:**
-
-1. **Data extraction** (cells 1-14): Query ChEMBL for drugs, targets, effects
-2. **Protein embeddings** (cells 15-18): Generate ESM-2 embeddings (GPU recommended, ~30 min)
-3. **Graph construction** (cells 19-35): Build NetworkX graph with nodes/edges
-4. **TransE training** (cell 36): Train embedding model (100 epochs, ~15 min on GPU)
-5. **Prediction** (cells 37-42): Generate novel drug-target and drug-effect predictions
-6. **Visualization** (cells 43-45): t-SNE plots, neighbor analysis
-
----
-
-## 📁 Project Structure
-
-```
-pharmacology-graph/
-├── code.ipynb                          # Main analysis notebook
-├── README.md                           # This file
-├── .gitignore                          # Git ignore rules
-├── requirements.txt                    # Python dependencies (UV)
-│
-├── chembl_36/
-│   └── chembl_36_sqlite/
-│       └── chembl_36.db               # ChEMBL database (not in git)
-│
-├── protein_nodes_with_embeddings.pkl  # ESM-2 protein embeddings (not in git)
-├── drug_nodes.pkl                     # Drug metadata (not in git)
-├── drug_effects.pkl                   # Drug-indication mappings (not in git)
-├── drug_protein_interactions.pkl      # Known drug-target edges (not in git)
-│
-├── graph_embeddings.npy               # Trained TransE embeddings (not in git)
-├── node_to_idx.npy                    # Node index mapping (not in git)
-│
-├── top_50_predicted_drug_protein.csv  # Novel target predictions
-├── top_50_predicted_drug_effects.csv  # Novel indication predictions
-└── drug_neighbors_visualization.png   # t-SNE cluster plot (not in git)
-```
-
----
-
-## 🔬 Technical Details
-
-### Model Architecture: TransE
-
-**TranslatingEmbeddings for Multi-Relational Graphs**
-
-For each edge `(head, relation, tail)`:
-- **Scoring function:** `f(h, r, t) = ||h + r - t||₂`
-- **Loss:** Margin-based ranking loss with negative sampling
-
-**Hyperparameters:**
-```python
-EMBEDDING_DIM = 128
-LEARNING_RATE = 0.01
-EPOCHS = 100
-BATCH_SIZE = 32
-MARGIN = 1.0
-```
-
-**Training stats:**
-- Nodes: ~1,400 (drugs: ~800, proteins: ~200, effects: ~400)
-- Edges: ~15,000 (drug-protein: ~12,000, drug-effect: ~3,000)
-- Training time: 15 minutes (NVIDIA GPU)
-- Final loss: ~0.15
-
-### ESM-2 Protein Embeddings
-
-**Model:** `facebook/esm2_t36_3B_UR50D`
-- Parameters: 3B
-- Embedding dimension: 2560
-- Context length: 1024 amino acids
-- Mean pooling over sequence length
+A static front-end. Prediction data is served from the project's GitHub Pages
+(<https://joevondahab.github.io/pharmacology-graph/>), so this Space and the Pages site share one
+source of truth.
 
-**Processing:**
-- Batch size: 4 proteins
-- Total proteins: ~200
-- Compute time: ~30 minutes (GPU) / ~3 hours (CPU)
-
-### Drug Representations
-
-**SMILES → Morgan Fingerprints**
-```python
-radius = 2
-n_bits = 2048
-```
-
-**Bridge to Graph Embeddings:**
-- Ridge regression: Fingerprint → TransE embedding
-- R² score: 0.67 (on training drugs)
-- Enables predictions for completely novel molecules
-
----
-
-## 🧪 Validation Strategy
-
-### 1. **Held-out Known Interactions**
-
-Split known drug-target edges:
-- Train: 80% (used for TransE)
-- Test: 20% (hidden during training)
-
-**Metrics:**
-- Recall@50: How many true targets appear in top-50 predictions?
-- Mean Reciprocal Rank (MRR)
-- AUROC for ranked predictions
-
-### 2. **Time-based Split**
-
-- Train on: Drug approvals ≤ 2015
-- Test on: Approvals > 2015
-- Simulates prospective prediction
-
-### 3. **Baseline Comparisons**
-
-| Method | Recall@50 | MRR | AUROC |
-|--------|-----------|-----|-------|
-| **Our Model (TransE)** | 0.68 | 0.42 | 0.83 |
-| ECFP Tanimoto (fingerprint) | 0.31 | 0.18 | 0.67 |
-| Random baseline | 0.02 | 0.01 | 0.50 |
-
-**Ablation studies:**
-- Without ESM-2 (random protein init): -0.15 AUROC
-- Without contrastive training: -0.11 AUROC
-
----
-
-## 📈 Use Cases
-
-### 1. **Drug Repurposing**
-
-Find new therapeutic uses for approved drugs:
-
-```python
-# Example: Query novel indications for Aspirin
-drug_smiles = "CC(=O)Oc1ccccc1C(=O)O"  # Aspirin
-predictions = predict_new_drug_embedding(drug_smiles, top_k=10)
-
-# Output: Predicted new effects beyond pain/inflammation
-# - Cardiovascular prevention (known)
-# - Colorectal cancer prevention (emerging evidence)
-```
-
-### 2. **Target Identification**
-
-Predict protein targets for experimental compounds:
-
-```python
-# Novel kinase inhibitor candidate
-novel_smiles = "Cc1ccc(Nc2nccc(...)...)cc1"
-targets = predict_protein_targets(novel_smiles, top_k=20)
-
-# Helps prioritize biochemical assays
-```
-
-### 3. **Off-target Prediction**
-
-Identify safety liabilities early:
-
-```python
-# Check for unintended receptor binding
-all_targets = predict_all_targets(drug_smiles, threshold=0.45)
-safety_flags = [t for t in all_targets if t in ['hERG', 'CYP3A4', 'Opioid']]
-```
-
----
-
-## 🎓 Scientific Contributions
-
-1. **Cross-modal contrastive learning** for drug discovery
-   - First work aligning SMILES + ESM-2 in shared space
-   
-2. **Pharmacology-aligned embeddings**
-   - Cluster by mechanism, not just structure
-   - Enables interpretable predictions
-
-3. **Scalable graph-based prediction**
-   - Handles multi-relational heterogeneous graphs
-   - GPU-accelerated inference (~1M predictions/sec)
-
-4. **Validated on real-world pharmacology**
-   - Recovers known drug classes (opioids, antivirals, statins)
-   - Predicts plausible novel targets with >80% precision
-
----
-
-## 📝 Citation
-
-```bibtex
-@software{pharmacology_graph_2025,
-  author = {Joe VonDahab},
-  title = {Pharmacology Knowledge Graph: Drug-Target-Effect Prediction},
-  year = {2025},
-  url = {https://github.com/JoeVonDahab/pharmacology-graph}
-}
-```
-
-**Related work:**
-- **TransE:** Bordes et al., "Translating Embeddings for Modeling Multi-relational Data" (NeurIPS 2013)
-- **ESM-2:** Lin et al., "Evolutionary-scale prediction of atomic-level protein structure" (Science 2023)
-- **ChEMBL:** Gaulton et al., "The ChEMBL database in 2017" (Nucleic Acids Research 2017)
-
----
-
-## 🤝 Contributing
-
-Contributions welcome! Areas for improvement:
-
-- [ ] Add attention-based graph neural network (GAT/RGCN)
-- [ ] Incorporate 3D protein structures (AlphaFold2)
-- [ ] Multi-task learning (toxicity + efficacy)
-- [ ] Temporal dynamics (drug resistance evolution)
-- [ ] Web interface (Streamlit/Gradio demo)
-
----
-
-## 📄 License
-
-MIT License - see LICENSE file for details
-
----
-
-## 🙏 Acknowledgments
-
-- **ChEMBL** for curated pharmacological data
-- **Meta AI** for ESM-2 protein language models
-- **RDKit** for cheminformatics tools
-- **PyTorch** ecosystem for deep learning infrastructure
-
----
-
-## 📧 Contact
-
-**Author:** Youssef Abo-Dahab
-**Repository:** [github.com/JoeVonDahab/pharmacology-graph](https://github.com/JoeVonDahab/pharmacology-graph)
-
-For questions or collaboration: [create an issue](https://github.com/JoeVonDahab/pharmacology-graph/issues)
-
----
+- Code & notebooks: <https://github.com/JoeVonDahab/pharmacology-graph>
+- Paper: <https://arxiv.org/abs/2603.01537>
 
-*Last updated: October 2025*
+Predictions are model outputs for research exploration only — not clinical or experimental validation.

SETUP.md

@@ -1,351 +0,0 @@
-# Setup Guide - Pharmacology Knowledge Graph
-
-Complete guide for setting up and running the Pharmacology Graph Explorer on your local machine.
-
----
-
-## 📋 Prerequisites
-
-- **Python**: 3.9 or higher
-- **RAM**: At least 8GB (16GB recommended for large datasets)
-- **Disk Space**: ~5GB for ChEMBL database and model files
-
----
-
-## 🚀 Quick Start (3 Steps)
-
-### 1. Clone the Repository
-
-```bash
-git clone https://github.com/JoeVonDahab/pharmacology-graph.git
-cd pharmacology-graph
-```
-
-### 2. Install Dependencies
-
-**Option A: Using `uv` (Recommended - Fast)**
-```bash
-# Install uv if you don't have it
-pip install uv
-
-# Install project dependencies
-uv pip install -r requirements_app.txt
-```
-
-**Option B: Using `pip`**
-```bash
-pip install -r requirements_app.txt
-```
-
-**Critical Version Note**: Make sure you have:
-- `gradio>=4.44.0` (not 4.36.x)
-- `numpy<2.0` (NumPy 2.x breaks compatibility)
-
-### 3. Run the App
-
-```bash
-./start_app.sh
-```
-
-Or manually:
-```bash
-python app.py
-```
-
-Then open your browser to: **http://localhost:7860**
-
----
-
-## 📦 What's Included
-
-### Essential Files (Required to Run)
-
-| File | Purpose | Size |
-|------|---------|------|
-| `app.py` | Main Gradio web application | 604 lines |
-| `requirements_app.txt` | Python dependencies | - |
-| `start_app.sh` | Convenience launcher script | - |
-
-### Data Files (Required to Run)
-
-All these CSV files are **required** for the app to work:
-
-| File | Description | Rows |
-|------|-------------|------|
-| `drug_nodes.csv` | Drug metadata (name, SMILES, ChEMBL ID) | 3,127 |
-| `protein_nodes_with_embeddings.csv` | Protein targets with ESM-2 features | 1,156 |
-| `drug_effects.csv` | Clinical effects and indications | 8,312 |
-| `drugs_interactions.csv` | Known drug-protein interactions | 11,703 |
-| `graph_embeddings.npy` | Trained TransE embeddings (128-dim) | 5,201 nodes |
-| `node_to_idx.npy` | Node ID mappings | - |
-| `top_50_predicted_drug_protein.csv` | Top predicted drug-target pairs | 50 |
-| `top_50_predicted_drug_effects.csv` | Top predicted drug-effect pairs | 50 |
-
-**Note**: The `.npy` files are NOT tracked in git (they're large). You need to generate them by running the notebook.
-
-### Documentation Files
-
-| File | Purpose |
-|------|---------|
-| `README.md` | Project overview and results |
-| `DEPLOYMENT.md` | Guide for deploying to Hugging Face Spaces |
-| `APP_GUIDE.md` | User guide for the web application |
-| `BUGFIX_SUMMARY.md` | Recent bug fixes (version compatibility) |
-
-### Development Files
-
-| File | Purpose |
-|------|---------|
-| `code.ipynb` | Main research notebook (data extraction, training, prediction) |
-| `requirements.txt` | Full dependencies for running the notebook |
-| `test_app.py` | Diagnostic test for app data loading |
-| `test_search.py` | Manual testing guide for search functionality |
-
----
-
-## 🔄 Full Setup (If You Want to Retrain the Model)
-
-If you want to regenerate the embeddings and predictions from scratch:
-
-### 1. Download ChEMBL Database
-
-```bash
-# The ChEMBL 36 database is large (~4.2GB)
-# It's already in chembl_36/chembl_36_sqlite/chembl_36.db
-# If missing, download from: https://ftp.ebi.ac.uk/pub/databases/chembl/ChEMBLdb/latest/
-```
-
-### 2. Install Full Dependencies
-
-```bash
-pip install -r requirements.txt
-```
-
-This includes:
-- `transformers` (for ESM-2 protein embeddings)
-- `torch` (for TransE training)
-- `rdkit` (for molecular fingerprints)
-- `sqlite3` (for ChEMBL database queries)
-
-### 3. Run the Notebook
-
-Open `code.ipynb` in Jupyter and run all cells:
-
-```bash
-jupyter notebook code.ipynb
-```
-
-This will:
-1. Extract data from ChEMBL (drugs, proteins, effects)
-2. Generate ESM-2 embeddings for proteins (~30 min on GPU)
-3. Train TransE graph model (100 epochs, ~10 min)
-4. Generate predictions using cosine similarity
-5. Export all CSV and NPY files
-
----
-
-## 🧪 Testing the App
-
-### Quick Test
-
-```bash
-python test_app.py
-```
-
-This checks:
-- ✓ All data files are present
-- ✓ Libraries are correctly installed
-- ✓ Search function works
-- ✓ Data structure is valid
-
-### Manual Testing
-
-1. Start the app: `./start_app.sh`
-2. Open: http://localhost:7860
-3. Search for: "Aspirin", "Morphine", or "Imatinib"
-4. Verify:
-   - Dropdown shows matching results
-   - Selecting a drug displays data tables
-   - Network visualization renders correctly
-   - No errors in console
-
----
-
-## 🐛 Troubleshooting
-
-### Port Already in Use
-
-```bash
-# Kill existing app instance
-pkill -f "python.*app.py"
-
-# Or use a different port
-GRADIO_SERVER_PORT=7861 python app.py
-```
-
-### Import Errors
-
-```bash
-# Check gradio version (must be >=4.44.0)
-pip show gradio
-
-# Check numpy version (must be <2.0)
-pip show numpy
-
-# Reinstall with correct versions
-pip install --force-reinstall "gradio>=4.44.0" "numpy<2.0"
-```
-
-### Missing Data Files
-
-If you see errors about missing CSV files:
-1. Make sure you cloned the full repository
-2. Run `code.ipynb` to regenerate data files
-3. Check `.gitignore` - some large files aren't tracked
-
-### Search Not Showing Results
-
-Make sure you:
-1. Updated to Gradio 4.44+ (older versions have bugs)
-2. Refreshed your browser after restarting the app
-3. Check console for debug output (should show "Search 'xxx' found N results")
-
----
-
-## 📊 Understanding the Output
-
-### When You Search for a Drug
-
-**Example: Searching "Aspirin"**
-
-You'll see:
-1. **Basic Info**: ChEMBL ID, SMILES structure
-2. **Known Targets**: Proteins with measured pChEMBL values
-3. **Predicted Targets**: Novel protein interactions (cosine similarity >0.7)
-4. **Clinical Effects**: Known indications and phases
-5. **Predicted Effects**: Potential repurposing opportunities
-6. **Network Graph**: Interactive visualization of relationships
-
-### Interpreting Predictions
-
-- **Similarity Score**: 0.0-1.0 (higher = more confident)
-  - >0.8: Very high confidence
-  - 0.7-0.8: High confidence
-  - 0.6-0.7: Moderate confidence
-  - <0.6: Low confidence (not shown)
-
-- **Confidence**: "Very High", "High", "Medium" based on similarity threshold
-
----
-
-## 🚢 Deploying to Hugging Face Spaces
-
-See `DEPLOYMENT.md` for complete step-by-step guide.
-
-**Quick version:**
-
-1. Create a Space at https://huggingface.co/spaces
-2. Upload these files:
-   - `app.py`
-   - All CSV files
-   - `graph_embeddings.npy` and `node_to_idx.npy`
-   - Rename `requirements_app.txt` → `requirements.txt`
-   - Use `SPACE_README.md` content for README.md
-3. Wait 2-3 minutes for build
-4. Your app will be live!
-
----
-
-## 🗑️ Unnecessary Files (Can Be Deleted)
-
-**For end users who just want to run the app:**
-
-You can safely delete:
-- `BUGFIX_SUMMARY.md` (development notes)
-- `APP_SUMMARY.md` (duplicate of APP_GUIDE.md)
-- `CHECKLIST.md` (deployment checklist)
-- `run_app.sh` (use `start_app.sh` instead)
-- `test_search.py` (manual testing guide, not needed if app works)
-- `requirements.txt` (only needed for notebook, use `requirements_app.txt`)
-
-**If you're NOT retraining the model:**
-- `code.ipynb` (the notebook)
-- `chembl_36/` directory (large database)
-- `full database.xml` (if present)
-
-**Keep these minimal files:**
-```
-pharmacology-graph/
-├── app.py
-├── requirements_app.txt
-├── start_app.sh
-├── test_app.py
-├── README.md
-├── DEPLOYMENT.md
-├── APP_GUIDE.md
-├── *.csv (all CSV files)
-├── graph_embeddings.npy
-└── node_to_idx.npy
-```
-
----
-
-## 📝 File Size Reference
-
-```
-drug_nodes.csv                          ~700 KB
-protein_nodes_with_embeddings.csv       ~400 KB
-drug_effects.csv                        ~800 KB
-drugs_interactions.csv                  ~1.5 MB
-graph_embeddings.npy                    ~5 MB
-node_to_idx.npy                         ~50 KB
-top_50_predicted_*.csv                  ~10 KB each
-```
-
-**Total app files**: ~10 MB  
-**With ChEMBL database**: ~4.2 GB  
-
----
-
-## 🤝 Contributing
-
-If you want to improve the project:
-
-1. Fork the repository
-2. Run the full notebook to understand the pipeline
-3. Make changes (better predictions, UI improvements, etc.)
-4. Test with `python test_app.py`
-5. Submit a pull request
-
----
-
-## 📚 Additional Resources
-
-- **ChEMBL Database**: https://www.ebi.ac.uk/chembl/
-- **ESM-2 Model**: https://github.com/facebookresearch/esm
-- **TransE Paper**: "Translating Embeddings for Modeling Multi-relational Data" (Bordes et al., 2013)
-- **Gradio Docs**: https://gradio.app/docs/
-
----
-
-## ⚖️ License
-
-This project uses publicly available ChEMBL data (CC BY-SA 3.0).  
-Model and code are provided as-is for research purposes only.
-
-**Not for clinical use.**
-
----
-
-## 📧 Support
-
-Issues? Questions?
-- Open an issue: https://github.com/JoeVonDahab/pharmacology-graph/issues
-- Check existing documentation: README.md, APP_GUIDE.md, DEPLOYMENT.md
-
----
-
-**Last Updated**: October 10, 2025  
-**Version**: 1.0.0  
-**Python**: 3.9+  
-**Gradio**: 4.44.1

SPACE_README.md

@@ -1,157 +0,0 @@
----
-title: Pharmacology Knowledge Graph Explorer
-emoji: 💊
-colorFrom: blue
-colorTo: green
-sdk: gradio
-sdk_version: 4.44.0
-app_file: app.py
-pinned: false
-license: mit
----
-
-# 💊 Pharmacology Knowledge Graph Explorer
-
-**Explore drug-target-effect relationships using AI-powered predictions**
-
-## 🎯 What is this?
-
-An interactive web app that lets you explore:
-- **Known interactions** between drugs and protein targets
-- **Predicted novel targets** for drug repurposing
-- **Clinical effects** and therapeutic indications
-- **Interactive network visualizations** of drug mechanisms
-
-## 🧬 How It Works
-
-This app uses a **TransE knowledge graph embedding model** trained on:
-- **800+ FDA-approved drugs** from ChEMBL database
-- **200+ human protein targets** with ESM-2 sequence embeddings
-- **400+ clinical effects** and disease indications
-
-The model learns a unified embedding space where:
-- Drugs with similar mechanisms cluster together
-- Predictions are based on cosine similarity in latent space
-- Novel drug-target pairs are ranked by confidence
-
-## 🔍 Example Queries
-
-Try searching for these drugs to see interesting results:
-
-| Drug | What you'll see |
-|------|----------------|
-| **Morphine** | Opioid receptor targets, analgesic effects, similar opioids (Fentanyl, Oxycodone) |
-| **Aspirin** | COX enzyme inhibition, anti-inflammatory effects, cardiovascular prevention |
-| **Metformin** | Diabetes targets, metabolic effects, potential repurposing for cancer/aging |
-| **Ibuprofen** | COX-1/COX-2 targets, NSAID effects, similar drugs (Naproxen, Diclofenac) |
-| **Talazoparib** | PARP family targets, cancer indications, predicted PARP homologs |
-
-## 📊 Model Performance
-
-**Validation metrics:**
-- **Precision@50:** ~90% pharmacologically coherent predictions
-- **Mean similarity (top predictions):** 0.48-0.60
-- **Baseline (random):** 0.23
-
-**Key achievements:**
-- ✅ Correctly clusters drugs by therapeutic class (e.g., opioids, antivirals, statins)
-- ✅ Recovers known drug-target relationships with high accuracy
-- ✅ Predicts plausible novel targets for experimental validation
-
-## 🕸️ Network Visualization Features
-
-The interactive graph shows:
-- 🔴 **Central drug node** (red)
-- 🔵 **Known protein targets** (blue, solid edges)
-- 🟢 **Predicted novel targets** (green, dashed edges)
-- 🟡 **Known clinical effects** (yellow, solid edges)
-- 🟨 **Predicted effects** (light yellow, dashed edges)
-
-**Controls:**
-- Toggle known/predicted interactions
-- Adjust max nodes displayed
-- Hover over nodes for details
-
-## 🚀 Use Cases
-
-1. **Drug Repurposing:** Find new therapeutic uses for existing drugs
-2. **Target Discovery:** Identify potential protein targets for lead compounds
-3. **Safety Assessment:** Predict off-target effects and side effects
-4. **Mechanism Exploration:** Understand drug action through network context
-
-## 📖 Citation
-
-If you use this tool in your research, please cite:
-
-```bibtex
-@software{pharmacology_graph_2025,
-  author = {Joe VonDahab},
-  title = {Pharmacology Knowledge Graph: Drug-Target-Effect Prediction},
-  year = {2025},
-  url = {https://github.com/JoeVonDahab/pharmacology-graph},
-  note = {Hugging Face Space: https://huggingface.co/spaces/YOUR_USERNAME/pharmacology-knowledge-graph}
-}
-```
-
-## 🔗 Links
-
-- **GitHub Repository:** [pharmacology-graph](https://github.com/JoeVonDahab/pharmacology-graph)
-- **Paper/Blog:** *(Coming soon)*
-- **Dataset:** [ChEMBL 36](https://www.ebi.ac.uk/chembl/)
-- **Protein Model:** [ESM-2 (Meta AI)](https://github.com/facebookresearch/esm)
-
-## ⚠️ Disclaimer
-
-**This is a research tool for exploratory analysis only.**
-
-- Predictions are computational and require experimental validation
-- Not intended for clinical decision-making or medical diagnosis
-- Drug repurposing candidates need rigorous testing before clinical use
-- Always consult scientific literature and regulatory databases
-
-## 🛠️ Technical Details
-
-**Model Architecture:**
-- **Graph embedding:** TransE (Translation-based)
-- **Embedding dimension:** 128
-- **Protein featurization:** ESM-2 (2560-dim) → mean pooling
-- **Drug featurization:** Morgan fingerprints (2048-bit)
-
-**Training:**
-- Epochs: 100
-- Learning rate: 0.01
-- Negative sampling: 1:1 ratio
-- Margin loss: 1.0
-
-**Data sources:**
-- ChEMBL 36 (drugs, targets, indications)
-- UniProt (protein sequences)
-- MeSH (clinical effects taxonomy)
-
-## 💡 Future Enhancements
-
-Planned features:
-- [ ] SMILES input for novel molecules
-- [ ] 3D protein structure integration (AlphaFold2)
-- [ ] Multi-task learning (toxicity + efficacy)
-- [ ] Batch analysis and export
-- [ ] API endpoint for programmatic access
-
-## 🤝 Contributing
-
-Found a bug or have a feature request? 
-- Open an issue on [GitHub](https://github.com/JoeVonDahab/pharmacology-graph/issues)
-- Submit a pull request
-- Contact: [create an issue](https://github.com/JoeVonDahab/pharmacology-graph/issues)
-
-## 📜 License
-
-MIT License - See [LICENSE](https://github.com/JoeVonDahab/pharmacology-graph/blob/main/LICENSE) for details
-
----
-
-**Built with:** 🤗 Gradio • PyTorch • NetworkX • Plotly • ESM-2
-
-**Powered by:** ChEMBL • Hugging Face Spaces
-
-*Last updated: October 2025*

USER_SETUP_SUMMARY.md

@@ -1,302 +0,0 @@
-# 📋 Summary: Setup for New Users
-
-## What Someone Needs to Do to Run Your App
-
-### **Method 1: Just Run the App (Recommended for Most Users)**
-
-```bash
-# 1. Clone repository
-git clone https://github.com/JoeVonDahab/pharmacology-graph.git
-cd pharmacology-graph
-
-# 2. Install dependencies
-pip install -r requirements_app.txt
-
-# 3. Download the .npy files (NOT in git)
-# They need to either:
-#   a) Run the notebook to generate them, OR
-#   b) Download from a release/separate link you provide
-
-# 4. Run the app
-./start_app.sh
-# Open browser to http://localhost:7860
-```
-
-**Requirements:**
-- Python 3.9+
-- 8GB RAM
-- ~10MB disk space
-
-**What they get:**
-- ✅ Interactive web app
-- ✅ Search 3,127 drugs
-- ✅ View predictions
-- ✅ Network visualization
-
----
-
-### **Method 2: Full Research Pipeline (Advanced Users)**
-
-```bash
-# 1-2. Same as above
-
-# 3. Download ChEMBL database (4.2GB)
-# Already included in repo if you pushed it, otherwise:
-# wget https://ftp.ebi.ac.uk/pub/databases/chembl/ChEMBLdb/releases/chembl_36/chembl_36_sqlite.tar.gz
-
-# 4. Install full dependencies
-pip install -r requirements.txt
-
-# 5. Run the notebook
-jupyter notebook code.ipynb
-# Run all cells (~40 minutes)
-
-# 6. Run the app
-./start_app.sh
-```
-
-**Requirements:**
-- Python 3.9+
-- 16GB RAM (for ESM-2)
-- GPU recommended (or 30+ min on CPU)
-- ~5GB disk space
-
-**What they get:**
-- ✅ Everything from Method 1
-- ✅ Ability to retrain model
-- ✅ Explore data processing
-- ✅ Modify predictions
-
----
-
-## 🗑️ Files You Can Remove Before Sharing
-
-Run this to clean up unnecessary files:
-
-```bash
-./cleanup.sh
-```
-
-Or manually remove:
-
-### Definitely Remove (Duplicates/Debugging)
-```bash
-rm -f BUGFIX_SUMMARY.md      # Development notes
-rm -f APP_SUMMARY.md          # Duplicate of APP_GUIDE.md
-rm -f CHECKLIST.md            # One-time deployment checklist
-rm -f run_app.sh              # Old script (use start_app.sh)
-rm -f test_search.py          # Manual testing guide
-rm -f main.py                 # Old experimental code
-rm -f .python-version         # Unnecessary version pin
-```
-
-### Optional Remove (Can Regenerate)
-```bash
-rm -f *.png                   # Visualization outputs
-rm -f *.pkl                   # Pickle files
-rm -f drug_mechanism_filtered.csv
-rm -f drug_warnings.csv
-rm -f edges_drug_protein.csv
-rm -f proteins_for_embedding.csv
-rm -f top_50_predicted_drug_targets.csv  # Duplicate
-```
-
-### Consider Removing (If Users Don't Need to Retrain)
-```bash
-rm -f code.ipynb              # Research notebook (750KB)
-rm -f requirements.txt        # Full deps (use requirements_app.txt only)
-rm -rf chembl_36/             # Database (4.2GB!)
-```
-
----
-
-## 📦 Minimal File Set for App-Only Distribution
-
-After cleanup, you'd have:
-
-```
-pharmacology-graph/
-├── app.py                                    # Main app
-├── requirements_app.txt                      # Dependencies
-├── start_app.sh                              # Launcher
-├── test_app.py                               # Diagnostic
-├── README.md                                 # Overview
-├── SETUP.md                                  # Setup guide ⭐
-├── FILES.md                                  # This file list
-├── DEPLOYMENT.md                             # HF Spaces guide
-├── APP_GUIDE.md                              # User guide
-├── SPACE_README.md                           # For HF deployment
-├── drug_nodes.csv                            # Data
-├── protein_nodes_with_embeddings.csv         # Data
-├── drug_effects.csv                          # Data
-├── drugs_interactions.csv                    # Data
-├── top_50_predicted_drug_protein.csv         # Predictions
-├── top_50_predicted_drug_effects.csv         # Predictions
-├── graph_embeddings.npy                      # Model ⚠️ NOT IN GIT
-├── node_to_idx.npy                           # Mapping ⚠️ NOT IN GIT
-└── .gitignore                                # Git config
-```
-
-**Total size**: ~10 MB (without .npy files)
-
----
-
-## ⚠️ IMPORTANT: The .npy Files Problem
-
-### The Issue
-The `.npy` files (`graph_embeddings.npy` and `node_to_idx.npy`) are:
-- **Essential** for the app to run
-- **NOT tracked in git** (they're in `.gitignore`)
-- **~5MB total** (not huge, but binary)
-
-### Solutions
-
-**Option 1: Add them to git** (Simplest)
-```bash
-# Edit .gitignore to allow these specific files
-echo '!graph_embeddings.npy' >> .gitignore
-echo '!node_to_idx.npy' >> .gitignore
-
-# Add and push
-git add graph_embeddings.npy node_to_idx.npy
-git commit -m "Add model embeddings"
-git push
-```
-
-**Option 2: GitHub Release** (Cleaner)
-1. Go to GitHub → Releases → Create new release
-2. Upload `graph_embeddings.npy` and `node_to_idx.npy`
-3. Update SETUP.md with download instructions:
-```bash
-# Download embeddings
-wget https://github.com/YOUR_USERNAME/pharmacology-graph/releases/download/v1.0/graph_embeddings.npy
-wget https://github.com/YOUR_USERNAME/pharmacology-graph/releases/download/v1.0/node_to_idx.npy
-```
-
-**Option 3: Git LFS** (Professional)
-```bash
-# Install Git LFS
-git lfs install
-
-# Track .npy files
-git lfs track "*.npy"
-git add .gitattributes
-
-# Add and push
-git add graph_embeddings.npy node_to_idx.npy
-git commit -m "Add model embeddings via LFS"
-git push
-```
-
-**Option 4: External Storage** (If files are huge)
-- Upload to Hugging Face Hub, Google Drive, Dropbox
-- Provide download link in SETUP.md
-
-### Recommendation
-For 5MB files, **Option 1** (just add to git) is fine. GitHub allows files up to 100MB.
-
----
-
-## 📝 Updated SETUP.md Instructions
-
-I've created **`SETUP.md`** with complete instructions including:
-
-✅ **Quick Start** (3 commands)  
-✅ **Full Setup** (retrain from scratch)  
-✅ **Troubleshooting** (version issues, port conflicts)  
-✅ **Testing** (verify it works)  
-✅ **Deployment** (Hugging Face Spaces)  
-✅ **File cleanup** (remove unnecessary files)
-
-**For new users, tell them:**
-> "Read **SETUP.md** for complete installation and usage instructions."
-
----
-
-## 🎯 Recommended Git Workflow
-
-### Before sharing on GitHub:
-
-1. **Clean up files**:
-   ```bash
-   ./cleanup.sh
-   ```
-
-2. **Add .npy files to git** (if you choose Option 1):
-   ```bash
-   echo '!graph_embeddings.npy' >> .gitignore
-   echo '!node_to_idx.npy' >> .gitignore
-   git add graph_embeddings.npy node_to_idx.npy .gitignore
-   git commit -m "Add model embeddings for app"
-   ```
-
-3. **Update README**:
-   - Already updated with Quick Start section
-   - Points to SETUP.md for details
-
-4. **Push everything**:
-   ```bash
-   git add .
-   git commit -m "Add setup documentation and cleanup scripts"
-   git push
-   ```
-
-### User experience:
-```bash
-git clone https://github.com/YOUR_USERNAME/pharmacology-graph
-cd pharmacology-graph
-pip install -r requirements_app.txt
-./start_app.sh
-```
-
-**That's it!** 3 commands and they're running.
-
----
-
-## 📊 File Organization Summary
-
-| Category | Files | Purpose | Size | In Git? |
-|----------|-------|---------|------|---------|
-| **App Core** | app.py, start_app.sh | Run the interface | 50KB | ✅ Yes |
-| **Data** | 6 CSV files | Drug/protein/effect data | 3.5MB | ✅ Yes |
-| **Model** | 2 NPY files | Trained embeddings | 5MB | ⚠️ **No** (need to add) |
-| **Predictions** | 2 CSV files | Top 50 predictions | 20KB | ✅ Yes |
-| **Docs** | 7 MD files | Setup, deployment, usage | 50KB | ✅ Yes |
-| **Research** | code.ipynb | Full pipeline | 500KB | ✅ Yes |
-| **Database** | chembl_36/ | ChEMBL SQLite | 4.2GB | ❌ No (too large) |
-| **Cleanup** | 10+ files | Duplicates, old versions | 2MB | ✅ Yes (but can remove) |
-
-**Action Items:**
-1. ✅ Created SETUP.md (complete guide)
-2. ✅ Created FILES.md (file manifest)
-3. ✅ Created cleanup.sh (remove junk)
-4. ✅ Updated README.md (quick start)
-5. ⚠️ **TODO**: Decide how to distribute .npy files (Option 1-4 above)
-
----
-
-## 🎓 Bottom Line
-
-**For someone to download and run your app, they need:**
-
-### Minimum (App Only):
-1. Clone repo
-2. `pip install -r requirements_app.txt`
-3. Download/generate `.npy` files
-4. `./start_app.sh`
-
-### Full (Research):
-1. Clone repo
-2. `pip install -r requirements.txt`
-3. Download ChEMBL (or use included)
-4. Run notebook → generates everything
-5. `./start_app.sh`
-
-**Unnecessary files removed**: ~15 files, saving 2MB and reducing clutter
-
-**New documentation**: 
-- ✅ SETUP.md (start here!)
-- ✅ FILES.md (what each file does)
-- ✅ cleanup.sh (automated cleanup)
-
-Users should read **SETUP.md** first. It has everything they need.

app.js

@@ -0,0 +1,233 @@
+'use strict';
+
+const V = 'p5';                       // bump on each data/app redeploy to bust caches
+// ponytail: data is served by the project's GitHub Pages (CORS: access-control-allow-origin *),
+// so this static Space stays ~25KB and shares one source of truth. Point at local 'data' if self-hosting.
+const DATA = 'https://joevondahab.github.io/pharmacology-graph/data';
+const bust = u => `${u}${u.includes('?') ? '&' : '?'}v=${V}`;
+const CHEMBL_CMP = id => `https://www.ebi.ac.uk/chembl/explore/compound/${id}`;
+const CHEMBL_TGT = id => `https://www.ebi.ac.uk/chembl/explore/target/${id}`;
+const MESH = id => `https://meshb.nlm.nih.gov/record/ui?ui=${id}`;
+const EXT = { drug: CHEMBL_CMP, protein: CHEMBL_TGT, mesh: MESH };
+
+// entity-type config for search + display
+const TYPES = {
+  drug:       { label: 'Drug',       chip: 'drug',  file: id => `predictions/${id}.json`,  index: 'drugs.json' },
+  indication: { label: 'Indication', chip: 'ind',   file: id => `by_indication/${id}.json`, index: 'indications.json' },
+  protein:    { label: 'Protein',    chip: 'prot',  file: id => `by_protein/${id}.json`,    index: 'proteins.json' },
+};
+
+let ENTITIES = [];     // combined search index: {type,id,name,...counts}
+let current = null;    // {type, view:[{key,label,linkType}], data}
+let activeList = null;
+let statusFilter = 'all';
+
+const $ = s => document.querySelector(s);
+const esc = s => String(s).replace(/[&<>"']/g, c => ({ '&': '&amp;', '<': '&lt;', '>': '&gt;', '"': '&quot;', "'": '&#39;' }[c]));
+
+async function init() {
+  try {
+    const [drugs, inds, prots, meta] = await Promise.all([
+      fetch(bust(`${DATA}/drugs.json`)).then(r => r.json()),
+      fetch(bust(`${DATA}/indications.json`)).then(r => r.json()),
+      fetch(bust(`${DATA}/proteins.json`)).then(r => r.json()),
+      fetch(bust(`${DATA}/meta.json`)).then(r => r.json()).catch(() => null),
+    ]);
+    ENTITIES = [
+      ...drugs.map(d => ({ ...d, type: 'drug' })),
+      ...inds.map(d => ({ ...d, type: 'indication' })),
+      ...prots.map(d => ({ ...d, type: 'protein' })),
+    ];
+    if (meta) {
+      $('#meta-line').textContent =
+        `${meta.num_drugs.toLocaleString()} drugs · ${meta.num_proteins.toLocaleString()} proteins · ` +
+        `${meta.num_indications.toLocaleString()} indications · top-${meta.top_k} each way`;
+      if (meta.metrics) {
+        const m = meta.metrics;
+        $('#metrics-note').textContent =
+          `Model test metrics — Drug–Protein PR-AUC ${m.dp.pr_auc.toFixed(3)}, Hits@10 ${m.dp['hits@10'].toFixed(3)} · ` +
+          `Drug–Indication PR-AUC ${m.di.pr_auc.toFixed(3)}, Hits@10 ${m.di['hits@10'].toFixed(3)}`;
+      }
+    }
+    buildExamples();
+    wireSearch();
+    if (location.hash.length > 1) {
+      const [t, id] = decodeURIComponent(location.hash.slice(1)).split(':');
+      if (id) select(t, id);
+    }
+  } catch (e) {
+    $('#empty-state').innerHTML = `<p style="color:#e06b6b">Failed to load data (${e}). Serve this folder over HTTP.</p>`;
+  }
+}
+
+function buildExamples() {
+  const picks = [
+    ['drug', 'SUNITINIB'], ['indication', 'Breast Neoplasms'],
+    ['drug', 'ASPIRIN'], ['indication', 'Hypertension'], ['protein', 'Histamine H1 receptor'],
+  ];
+  const box = $('#examples');
+  picks.forEach(([type, name]) => {
+    const e = ENTITIES.find(x => x.type === type && x.name.toUpperCase() === name.toUpperCase());
+    if (!e) return;
+    const b = document.createElement('button');
+    b.innerHTML = `<span class="chip ${TYPES[type].chip}">${TYPES[type].label}</span> ${esc(e.name)}`;
+    b.onclick = () => select(type, e.id);
+    box.appendChild(b);
+  });
+}
+
+// ---------- search ----------
+let activeIdx = -1, matches = [];
+function wireSearch() {
+  const input = $('#search'), list = $('#suggestions');
+  input.addEventListener('input', () => {
+    const q = input.value.trim().toLowerCase();
+    if (!q) { list.hidden = true; return; }
+    matches = ENTITIES
+      .filter(e => e.name.toLowerCase().includes(q) || e.id.toLowerCase().includes(q))
+      .sort((a, b) => (a.name.toLowerCase().indexOf(q) - b.name.toLowerCase().indexOf(q)) || a.name.length - b.name.length)
+      .slice(0, 14);
+    renderSuggestions(q);
+  });
+  input.addEventListener('keydown', e => {
+    if (list.hidden) return;
+    if (e.key === 'ArrowDown') { activeIdx = Math.min(activeIdx + 1, matches.length - 1); paint(); e.preventDefault(); }
+    else if (e.key === 'ArrowUp') { activeIdx = Math.max(activeIdx - 1, 0); paint(); e.preventDefault(); }
+    else if (e.key === 'Enter') { const m = matches[activeIdx] || matches[0]; if (m) select(m.type, m.id); }
+    else if (e.key === 'Escape') list.hidden = true;
+  });
+  document.addEventListener('click', e => { if (!e.target.closest('.search-box')) list.hidden = true; });
+}
+function renderSuggestions(q) {
+  const list = $('#suggestions'); activeIdx = -1;
+  if (!matches.length) { list.hidden = true; return; }
+  list.innerHTML = matches.map((e, i) => {
+    const t = TYPES[e.type];
+    const sub = e.type === 'drug' ? `${e.kp}+${e.ki} known` : `${e.kd} known drugs`;
+    return `<li data-i="${i}"><span class="s-left"><span class="chip ${t.chip}">${t.label}</span>
+      <span class="sname">${hl(e.name, q)}</span></span><span class="sid">${e.id} · ${sub}</span></li>`;
+  }).join('');
+  [...list.children].forEach((li, i) => li.onclick = () => select(matches[i].type, matches[i].id));
+  list.hidden = false;
+}
+function paint() { [...$('#suggestions').children].forEach(li => li.classList.toggle('active', +li.dataset.i === activeIdx)); }
+function hl(text, q) {
+  const i = text.toLowerCase().indexOf(q);
+  return i < 0 ? esc(text) : esc(text.slice(0, i)) + '<mark>' + esc(text.slice(i, i + q.length)) + '</mark>' + esc(text.slice(i + q.length));
+}
+
+// ---------- selection ----------
+async function select(type, id) {
+  $('#suggestions').hidden = true; $('#search').value = '';
+  let data;
+  try {
+    data = await fetch(bust(`${DATA}/${TYPES[type].file(id)}`)).then(r => { if (!r.ok) throw new Error(r.status); return r.json(); });
+  } catch (e) { alert(`No predictions found for ${type} "${id}".`); return; }
+
+  let view, head;
+  if (type === 'drug') {
+    view = [{ key: 'proteins', label: 'Protein targets', linkType: 'protein' },
+            { key: 'indications', label: 'Indications', linkType: 'mesh' }];
+    head = { name: data.drug.name, id: data.drug.id, ext: 'drug' };
+  } else if (type === 'indication') {
+    view = [{ key: 'drugs', label: 'Predicted drugs', linkType: 'drug' }];
+    head = { name: data.indication.name, id: data.indication.id, ext: 'mesh' };
+  } else {
+    view = [{ key: 'drugs', label: 'Predicted drugs', linkType: 'drug' }];
+    head = { name: data.protein.name, id: data.protein.id, ext: 'protein' };
+  }
+  current = { type, view, data, head };
+  activeList = view[0].key; statusFilter = 'all';
+  history.replaceState(null, '', `#${type}:${id}`);
+  $('#empty-state').hidden = true; $('#result').hidden = false;
+  renderHead(); renderControls(); renderTable();
+}
+
+function renderHead() {
+  const { type, head, view, data } = current;
+  $('#drug-name').textContent = head.name;
+  $('#type-chip').className = `chip ${TYPES[type].chip}`;
+  $('#type-chip').textContent = TYPES[type].label;
+  const link = $('#drug-link'); link.href = EXT[head.ext](head.id); link.textContent = `${head.id} ↗`;
+
+  const cards = [];
+  const count = (key, st) => (data[key] || []).filter(r => r.status === st).length;
+  if (type === 'drug') {
+    cards.push(['known', count('proteins', 'known'), 'known prot'], ['novel', count('proteins', 'novel'), 'novel prot'],
+               ['known', count('indications', 'known'), 'known ind'], ['novel', count('indications', 'novel'), 'novel ind']);
+  } else {
+    cards.push(['known', count('drugs', 'known'), 'known drugs'], ['novel', count('drugs', 'novel'), 'novel drugs']);
+  }
+  $('#stat-cards').innerHTML = cards.map(([c, n, l]) => `<div class="stat ${c}"><b>${n}</b><span>${l}</span></div>`).join('');
+}
+
+function renderControls() {
+  const tabs = $('#tabs');
+  if (current.view.length > 1) {
+    tabs.hidden = false;
+    tabs.innerHTML = current.view.map(v =>
+      `<button class="tab ${v.key === activeList ? 'active' : ''}" data-list="${v.key}">${v.label}</button>`).join('');
+  } else { tabs.hidden = true; }
+}
+
+function activeRows() {
+  const rows = current.data[activeList] || [];
+  const q = $('#table-filter').value.trim().toLowerCase();
+  return rows.filter(r => statusFilter === 'all' || r.status === statusFilter)
+            .filter(r => !q || r.name.toLowerCase().includes(q) || r.id.toLowerCase().includes(q));
+}
+
+function renderTable() {
+  const v = current.view.find(x => x.key === activeList);
+  const linkFn = EXT[v.linkType];
+  const colHead = v.linkType === 'drug' ? 'Drug' : (v.linkType === 'protein' ? 'Target' : 'Indication');
+  $('#th-target').textContent = colHead;
+  const rows = activeRows();
+  $('#pred-body').innerHTML = rows.map(r => `
+    <tr>
+      <td class="c-rank">${r.rank}</td>
+      <td class="tname">${esc(r.name)}</td>
+      <td class="tid"><a href="${linkFn(r.id)}" target="_blank" rel="noopener">${r.id} ↗</a></td>
+      <td><div class="scorewrap"><div class="bar"><i style="width:${Math.round(r.score * 100)}%"></i></div>
+        <span class="scoreval">${r.score.toFixed(3)}</span></div></td>
+      <td><span class="badge ${r.status}">${r.status}</span></td>
+    </tr>`).join('');
+  $('#no-rows').hidden = rows.length > 0;
+}
+
+// ---------- controls ----------
+document.addEventListener('click', e => {
+  const t = e.target;
+  if (t.classList.contains('tab')) { activeList = t.dataset.list; renderControls(); renderTable(); }
+  if (t.classList.contains('chip') && t.dataset.status) {
+    statusFilter = t.dataset.status;
+    document.querySelectorAll('#status-filter .chip').forEach(x => x.classList.toggle('active', x === t));
+    renderTable();
+  }
+  if (t.id === 'dl-csv') downloadCSV();
+  if (t.id === 'dl-json') downloadJSON();
+});
+document.addEventListener('input', e => { if (e.target.id === 'table-filter') renderTable(); });
+
+// ---------- downloads ----------
+function downloadCSV() {
+  if (!current) return;
+  const v = current.view.find(x => x.key === activeList);
+  const what = v.linkType === 'drug' ? 'drug' : (v.linkType === 'protein' ? 'protein' : 'indication');
+  const head = ['rank', `${what}_id`, `${what}_name`, 'score', 'raw_score', 'status'];
+  const lines = [head.join(',')];
+  for (const r of activeRows())
+    lines.push([r.rank, r.id, `"${r.name.replace(/"/g, '""')}"`, r.score, r.raw ?? '', r.status].join(','));
+  save(lines.join('\n'), `${current.head.id}_${activeList}.csv`, 'text/csv');
+}
+function downloadJSON() {
+  if (!current) return;
+  save(JSON.stringify(current.data, null, 2), `${current.head.id}_predictions.json`, 'application/json');
+}
+function save(text, name, type) {
+  const url = URL.createObjectURL(new Blob([text], { type }));
+  const a = document.createElement('a'); a.href = url; a.download = name; a.click();
+  setTimeout(() => URL.revokeObjectURL(url), 1000);
+}
+
+init();

app.py

@@ -1,609 +0,0 @@
-import gradio as gr
-import pandas as pd
-import numpy as np
-import plotly.graph_objects as go
-import networkx as nx
-from sklearn.metrics.pairwise import cosine_similarity
-import json
-
-# Load data
-print("Loading data...")
-drug_nodes = pd.read_csv('drug_nodes.csv')
-protein_nodes = pd.read_csv('protein_nodes_with_embeddings.csv')
-drug_effects = pd.read_csv('drug_effects.csv')
-drugs_interactions = pd.read_csv('drugs_interactions.csv')
-
-# Load predictions
-predicted_targets = pd.read_csv('top_50_predicted_drug_protein.csv')
-predicted_effects = pd.read_csv('top_50_predicted_drug_effects.csv')
-
-# Load embeddings
-embeddings = np.load('graph_embeddings.npy')
-node_to_idx = np.load('node_to_idx.npy', allow_pickle=True).item()
-
-print(f"✓ Loaded {len(drug_nodes)} drugs, {len(protein_nodes)} proteins, {len(drug_effects)} effects")
-
-# Create reverse mappings
-internal_to_drug_info = dict(zip(drug_nodes['drug_internal_id'], 
-                                 zip(drug_nodes['drug_id'], drug_nodes['drug_name'])))
-protein_id_to_name = dict(zip(protein_nodes['protein_id'], protein_nodes['protein_name']))
-effect_id_to_name = dict(zip(drug_effects['effect_id'], drug_effects['effect_name']))
-
-def search_drugs(query):
-    """Search for drugs by name or ChEMBL ID"""
-    try:
-        if not query:
-            return []
-        
-        query = query.lower()
-        matches = drug_nodes[
-            drug_nodes['drug_name'].str.lower().str.contains(query, na=False) |
-            drug_nodes['drug_id'].str.lower().str.contains(query, na=False)
-        ]
-        
-        # Return list of tuples (display_name, drug_internal_id)
-        results = [(f"{row['drug_name']} ({row['drug_id']})", row['drug_internal_id']) 
-                   for _, row in matches.head(20).iterrows()]
-        
-        return results
-    except Exception as e:
-        print(f"Error in search_drugs: {e}")
-        return []
-
-def get_drug_info(drug_internal_id):
-    """Get basic drug information"""
-    if drug_internal_id is None:
-        return "No drug selected"
-    
-    drug_row = drug_nodes[drug_nodes['drug_internal_id'] == drug_internal_id]
-    if len(drug_row) == 0:
-        return "Drug not found"
-    
-    drug_row = drug_row.iloc[0]
-    
-    info = f"""
-## 💊 {drug_row['drug_name']}
-
-**ChEMBL ID:** {drug_row['drug_id']}  
-**SMILES:** `{drug_row['smile'][:100]}...`
-"""
-    return info
-
-def get_known_targets(drug_internal_id):
-    """Get known protein targets for a drug"""
-    if drug_internal_id is None:
-        return pd.DataFrame()
-    
-    # Get ChEMBL ID
-    drug_row = drug_nodes[drug_nodes['drug_internal_id'] == drug_internal_id]
-    if len(drug_row) == 0:
-        return pd.DataFrame()
-    
-    drug_chembl_id = drug_row.iloc[0]['drug_id']
-    
-    # Get known interactions
-    known = drugs_interactions[drugs_interactions['drug_id'] == drug_chembl_id].copy()
-    
-    if len(known) == 0:
-        return pd.DataFrame(columns=['Target Protein', 'pChEMBL (max)', 'Measurements'])
-    
-    # Format for display
-    display_df = pd.DataFrame({
-        'Target Protein': known['protein_name'].values,
-        'ChEMBL ID': known['protein_id'].values,
-        'pChEMBL (max)': known['pchembl_max'].values,
-        'pChEMBL (avg)': known['pchembl_avg'].values,
-        'Measurements': known['num_measurements'].values
-    })
-    
-    return display_df.sort_values('pChEMBL (max)', ascending=False)
-
-def get_known_effects(drug_internal_id):
-    """Get known clinical effects for a drug"""
-    if drug_internal_id is None:
-        return pd.DataFrame()
-    
-    # Get known effects
-    known = drug_effects[drug_effects['drug_internal_id'] == drug_internal_id].copy()
-    
-    if len(known) == 0:
-        return pd.DataFrame(columns=['Clinical Effect', 'Phase', 'References'])
-    
-    # Format for display
-    display_df = pd.DataFrame({
-        'Clinical Effect': known['effect_name'].values,
-        'MeSH ID': known['effect_id'].values,
-        'Phase': known['indication_phase'].values,
-        'References': known['num_references'].values
-    })
-    
-    return display_df.sort_values('Phase', ascending=False)
-
-def get_predicted_targets(drug_internal_id, top_k=10):
-    """Get predicted novel protein targets"""
-    if drug_internal_id is None:
-        return pd.DataFrame()
-    
-    # Get ChEMBL ID
-    drug_row = drug_nodes[drug_nodes['drug_internal_id'] == drug_internal_id]
-    if len(drug_row) == 0:
-        return pd.DataFrame()
-    
-    drug_chembl_id = drug_row.iloc[0]['drug_id']
-    
-    # Get known targets to filter out
-    known_targets = set(drugs_interactions[
-        drugs_interactions['drug_id'] == drug_chembl_id
-    ]['protein_id'].values)
-    
-    # Check if drug is in embedding space
-    if drug_internal_id not in node_to_idx:
-        return pd.DataFrame(columns=['Predicted Target', 'Similarity', 'Confidence'])
-    
-    # Get drug embedding
-    drug_idx = node_to_idx[drug_internal_id]
-    drug_emb = embeddings[drug_idx].reshape(1, -1)
-    
-    # Compute similarities to all proteins
-    predictions = []
-    for _, protein_row in protein_nodes.iterrows():
-        protein_id = protein_row['protein_id']
-        protein_name = protein_row['protein_name']
-        
-        # Skip known targets
-        if protein_id in known_targets:
-            continue
-        
-        # Get protein embedding
-        if protein_id not in node_to_idx:
-            continue
-        
-        protein_idx = node_to_idx[protein_id]
-        protein_emb = embeddings[protein_idx].reshape(1, -1)
-        
-        # Calculate similarity
-        sim = cosine_similarity(drug_emb, protein_emb)[0][0]
-        
-        predictions.append({
-            'Predicted Target': protein_name,
-            'ChEMBL ID': protein_id,
-            'Similarity': f"{sim:.4f}",
-            'Confidence': 'High' if sim > 0.5 else 'Medium' if sim > 0.45 else 'Low'
-        })
-    
-    # Sort and return top-k
-    predictions_df = pd.DataFrame(predictions)
-    if len(predictions_df) == 0:
-        return pd.DataFrame(columns=['Predicted Target', 'Similarity', 'Confidence'])
-    
-    predictions_df['Similarity_Float'] = predictions_df['Similarity'].astype(float)
-    predictions_df = predictions_df.sort_values('Similarity_Float', ascending=False).head(top_k)
-    
-    return predictions_df[['Predicted Target', 'ChEMBL ID', 'Similarity', 'Confidence']]
-
-def get_predicted_effects(drug_internal_id, top_k=10):
-    """Get predicted novel clinical effects"""
-    if drug_internal_id is None:
-        return pd.DataFrame()
-    
-    # Get known effects to filter out
-    known_effects = set(drug_effects[
-        drug_effects['drug_internal_id'] == drug_internal_id
-    ]['effect_id'].values)
-    
-    # Check if drug is in embedding space
-    if drug_internal_id not in node_to_idx:
-        return pd.DataFrame(columns=['Predicted Effect', 'Similarity', 'Confidence'])
-    
-    # Get drug embedding
-    drug_idx = node_to_idx[drug_internal_id]
-    drug_emb = embeddings[drug_idx].reshape(1, -1)
-    
-    # Compute similarities to all effects
-    predictions = []
-    unique_effects = drug_effects[['effect_id', 'effect_name']].drop_duplicates()
-    
-    for _, effect_row in unique_effects.iterrows():
-        effect_id = effect_row['effect_id']
-        effect_name = effect_row['effect_name']
-        
-        # Skip known effects
-        if effect_id in known_effects:
-            continue
-        
-        # Get effect embedding
-        if effect_id not in node_to_idx:
-            continue
-        
-        effect_idx = node_to_idx[effect_id]
-        effect_emb = embeddings[effect_idx].reshape(1, -1)
-        
-        # Calculate similarity
-        sim = cosine_similarity(drug_emb, effect_emb)[0][0]
-        
-        predictions.append({
-            'Predicted Effect': effect_name,
-            'MeSH ID': effect_id,
-            'Similarity': f"{sim:.4f}",
-            'Confidence': 'High' if sim > 0.5 else 'Medium' if sim > 0.45 else 'Low'
-        })
-    
-    # Sort and return top-k
-    predictions_df = pd.DataFrame(predictions)
-    if len(predictions_df) == 0:
-        return pd.DataFrame(columns=['Predicted Effect', 'Similarity', 'Confidence'])
-    
-    predictions_df['Similarity_Float'] = predictions_df['Similarity'].astype(float)
-    predictions_df = predictions_df.sort_values('Similarity_Float', ascending=False).head(top_k)
-    
-    return predictions_df[['Predicted Effect', 'MeSH ID', 'Similarity', 'Confidence']]
-
-def create_network_visualization(drug_internal_id, show_known=True, show_predicted=True, max_nodes=20):
-    """Create interactive network visualization using Plotly"""
-    if drug_internal_id is None:
-        return None
-    
-    # Get drug info
-    drug_row = drug_nodes[drug_nodes['drug_internal_id'] == drug_internal_id]
-    if len(drug_row) == 0:
-        return None
-    
-    drug_name = drug_row.iloc[0]['drug_name']
-    drug_chembl_id = drug_row.iloc[0]['drug_id']
-    
-    # Create network
-    G = nx.Graph()
-    
-    # Add central drug node
-    G.add_node(drug_name, node_type='drug', color='#FF6B6B', size=30)
-    
-    edge_traces = []
-    node_texts = []
-    
-    # Add known targets
-    if show_known:
-        known_targets_df = get_known_targets(drug_internal_id)
-        for idx, row in known_targets_df.head(max_nodes // 2).iterrows():
-            target_name = row['Target Protein'][:30]
-            G.add_node(target_name, node_type='protein_known', color='#4ECDC4', size=15)
-            G.add_edge(drug_name, target_name, edge_type='known', color='#95E1D3', width=3)
-    
-    # Add predicted targets
-    if show_predicted:
-        predicted_targets_df = get_predicted_targets(drug_internal_id, top_k=max_nodes // 2)
-        for idx, row in predicted_targets_df.iterrows():
-            target_name = row['Predicted Target'][:30]
-            similarity = float(row['Similarity'])
-            G.add_node(target_name, node_type='protein_predicted', color='#A8E6CF', size=10)
-            G.add_edge(drug_name, target_name, edge_type='predicted', 
-                      color='#DCEDC8', width=1.5, similarity=similarity)
-    
-    # Add known effects
-    if show_known:
-        known_effects_df = get_known_effects(drug_internal_id)
-        for idx, row in known_effects_df.head(max_nodes // 3).iterrows():
-            effect_name = row['Clinical Effect'][:30]
-            G.add_node(effect_name, node_type='effect_known', color='#FFD93D', size=15)
-            G.add_edge(drug_name, effect_name, edge_type='known_effect', color='#FCF4A3', width=3)
-    
-    # Add predicted effects
-    if show_predicted:
-        predicted_effects_df = get_predicted_effects(drug_internal_id, top_k=max_nodes // 3)
-        for idx, row in predicted_effects_df.iterrows():
-            effect_name = row['Predicted Effect'][:30]
-            similarity = float(row['Similarity'])
-            G.add_node(effect_name, node_type='effect_predicted', color='#FFF9A3', size=10)
-            G.add_edge(drug_name, effect_name, edge_type='predicted_effect', 
-                      color='#FFFACD', width=1.5, similarity=similarity)
-    
-    # Calculate layout
-    pos = nx.spring_layout(G, k=2, iterations=50, seed=42)
-    
-    # Create edge traces
-    edge_traces = []
-    for edge in G.edges(data=True):
-        x0, y0 = pos[edge[0]]
-        x1, y1 = pos[edge[1]]
-        
-        edge_trace = go.Scatter(
-            x=[x0, x1, None],
-            y=[y0, y1, None],
-            mode='lines',
-            line=dict(
-                width=edge[2].get('width', 1),
-                color=edge[2].get('color', '#888')
-            ),
-            hoverinfo='none',
-            showlegend=False
-        )
-        edge_traces.append(edge_trace)
-    
-    # Create node trace
-    node_x = []
-    node_y = []
-    node_color = []
-    node_size = []
-    node_text = []
-    
-    for node in G.nodes(data=True):
-        x, y = pos[node[0]]
-        node_x.append(x)
-        node_y.append(y)
-        node_color.append(node[1].get('color', '#888'))
-        node_size.append(node[1].get('size', 10))
-        
-        # Create hover text
-        node_type = node[1].get('node_type', 'unknown')
-        if node_type == 'drug':
-            text = f"<b>{node[0]}</b><br>Type: Drug<br>ChEMBL: {drug_chembl_id}"
-        elif 'protein' in node_type:
-            text = f"<b>{node[0]}</b><br>Type: Protein Target<br>{'Known' if 'known' in node_type else 'Predicted'}"
-        elif 'effect' in node_type:
-            text = f"<b>{node[0]}</b><br>Type: Clinical Effect<br>{'Known' if 'known' in node_type else 'Predicted'}"
-        else:
-            text = f"<b>{node[0]}</b>"
-        
-        node_text.append(text)
-    
-    node_trace = go.Scatter(
-        x=node_x,
-        y=node_y,
-        mode='markers+text',
-        text=[node[0] for node in G.nodes()],
-        textposition='top center',
-        textfont=dict(size=8),
-        hovertext=node_text,
-        hoverinfo='text',
-        marker=dict(
-            size=node_size,
-            color=node_color,
-            line=dict(width=2, color='white')
-        ),
-        showlegend=False
-    )
-    
-    # Create figure
-    fig = go.Figure(data=edge_traces + [node_trace])
-    
-    fig.update_layout(
-        title=dict(
-            text=f"Knowledge Graph for {drug_name}",
-            font=dict(size=20)
-        ),
-        showlegend=False,
-        hovermode='closest',
-        margin=dict(b=0, l=0, r=0, t=40),
-        xaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
-        yaxis=dict(showgrid=False, zeroline=False, showticklabels=False),
-        plot_bgcolor='rgba(240,240,240,0.9)',
-        height=700
-    )
-    
-    # Add legend manually
-    fig.add_annotation(
-        text="<b>Legend:</b><br>" +
-             "🔴 Drug | 🔵 Known Target | 🟢 Predicted Target<br>" +
-             "🟡 Known Effect | 🟨 Predicted Effect",
-        xref="paper", yref="paper",
-        x=0.02, y=0.98,
-        showarrow=False,
-        bgcolor="white",
-        bordercolor="black",
-        borderwidth=1,
-        font=dict(size=10),
-        align="left"
-    )
-    
-    return fig
-
-def analyze_drug(search_query, drug_selection, show_known, show_predicted, max_nodes):
-    """Main analysis function"""
-    try:
-        if drug_selection is None:
-            return (
-                "Please search and select a drug from the dropdown",
-                pd.DataFrame(),
-                pd.DataFrame(),
-                pd.DataFrame(),
-                pd.DataFrame(),
-                None
-            )
-        
-        # Extract drug_internal_id from selection
-        drug_internal_id = drug_selection
-        
-        # Get all data with error handling
-        info = get_drug_info(drug_internal_id)
-        known_targets = get_known_targets(drug_internal_id)
-        known_effects = get_known_effects(drug_internal_id)
-        pred_targets = get_predicted_targets(drug_internal_id, top_k=15)
-        pred_effects = get_predicted_effects(drug_internal_id, top_k=15)
-        network = create_network_visualization(drug_internal_id, show_known, show_predicted, max_nodes)
-        
-        return info, known_targets, known_effects, pred_targets, pred_effects, network
-    
-    except Exception as e:
-        error_msg = f"⚠️ Error analyzing drug: {str(e)}\n\nPlease try selecting a different drug or check the console for details."
-        print(f"Error in analyze_drug: {e}")
-        import traceback
-        traceback.print_exc()
-        
-        return (
-            error_msg,
-            pd.DataFrame(),
-            pd.DataFrame(),
-            pd.DataFrame(),
-            pd.DataFrame(),
-            None
-        )
-
-# Create Gradio interface
-with gr.Blocks(title="Pharmacology Knowledge Graph Explorer", theme=gr.themes.Soft()) as demo:
-    gr.Markdown(
-        """
-        # 💊 Pharmacology Knowledge Graph Explorer
-        
-        Explore drug-target-effect relationships using AI-powered predictions from a TransE knowledge graph model.
-        
-        **Features:**
-        - 🔍 Search **2,526** FDA-approved drugs
-        - 🎯 View known and predicted protein targets
-        - 💉 Discover potential therapeutic uses (drug repurposing)
-        - 🕸️ Interactive network visualization
-        
-        ---
-        """
-    )
-    
-    with gr.Row():
-        with gr.Column(scale=1):
-            gr.Markdown("### 🔍 Drug Search")
-            search_box = gr.Textbox(
-                label="Search by drug name or ChEMBL ID",
-                placeholder="e.g., Aspirin, Morphine, CHEMBL25",
-                info="Type to search, then select from dropdown"
-            )
-            
-            drug_dropdown = gr.Dropdown(
-                label="Select Drug",
-                choices=[],
-                interactive=True
-            )
-            
-            gr.Markdown("### ⚙️ Visualization Settings")
-            show_known_checkbox = gr.Checkbox(label="Show known interactions", value=True)
-            show_predicted_checkbox = gr.Checkbox(label="Show predicted interactions", value=True)
-            max_nodes_slider = gr.Slider(
-                minimum=10, maximum=50, value=20, step=5,
-                label="Max nodes to display"
-            )
-            
-            analyze_btn = gr.Button("🔬 Analyze Drug", variant="primary", size="lg")
-        
-        with gr.Column(scale=2):
-            drug_info = gr.Markdown("### Select a drug to begin analysis")
-    
-    gr.Markdown("---")
-    
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown("### 🎯 Known Protein Targets")
-            known_targets_table = gr.Dataframe(
-                headers=['Target Protein', 'ChEMBL ID', 'pChEMBL (max)', 'pChEMBL (avg)', 'Measurements'],
-                interactive=False
-            )
-        
-        with gr.Column():
-            gr.Markdown("### 🔮 Predicted Novel Targets")
-            predicted_targets_table = gr.Dataframe(
-                headers=['Predicted Target', 'ChEMBL ID', 'Similarity', 'Confidence'],
-                interactive=False
-            )
-    
-    gr.Markdown("---")
-    
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown("### 💉 Known Clinical Effects")
-            known_effects_table = gr.Dataframe(
-                headers=['Clinical Effect', 'MeSH ID', 'Phase', 'References'],
-                interactive=False
-            )
-        
-        with gr.Column():
-            gr.Markdown("### 💡 Predicted Novel Effects (Repurposing)")
-            predicted_effects_table = gr.Dataframe(
-                headers=['Predicted Effect', 'MeSH ID', 'Similarity', 'Confidence'],
-                interactive=False
-            )
-    
-    gr.Markdown("---")
-    
-    gr.Markdown("### 🕸️ Interactive Knowledge Graph")
-    network_plot = gr.Plot()
-    
-    gr.Markdown(
-        """
-        ---
-        
-        ### 📊 About the Model
-        
-        This app uses a **TransE knowledge graph embedding model** trained on:
-        - **2,526** FDA-approved drugs
-        - **1,156** human protein targets (with ESM-2 embeddings)
-        - **1,065** clinical indications
-        - **18,789** total known links
-        
-        **Prediction method:** Cosine similarity in learned embedding space  
-        **Model performance:** ~90% precision on top-50 predictions
-        
-        **By:** Youssef Abo-Dahab, Pharm.D.  
-        **AICD3 M.S. Candidate, UCSF** | October 2025
-        
-        [GitHub Repository](https://github.com/JoeVonDahab/pharmacology-graph)
-        
-        ---
-        **Disclaimer:** This is a research tool for exploratory analysis only. Predictions should be validated experimentally. Not for clinical use.
-        """
-    )
-    
-    # Event handlers
-    # Store mapping globally to avoid re-searching
-    drug_display_to_id = gr.State({})
-    selected_drug_id = gr.State()
-    
-    def update_dropdown(search_query):
-        if not search_query or len(search_query) < 1:
-            return gr.update(choices=[]), {}
-        
-        results = search_drugs(search_query)
-        print(f"Search '{search_query}' found {len(results)} results")  # Debug output
-        
-        # Create mapping: display_name -> internal_id
-        mapping = {display: internal_id for display, internal_id in results}
-        choices = list(mapping.keys())
-        
-        print(f"Returning choices: {choices[:3]}..." if len(choices) > 3 else f"Returning choices: {choices}")  # Debug
-        return gr.update(choices=choices, value=None), mapping
-    
-    def store_selection(dropdown_value, mapping):
-        if not dropdown_value or not mapping:
-            return None
-        
-        # Get internal_id from mapping
-        return mapping.get(dropdown_value, None)
-    
-    search_box.change(
-        fn=update_dropdown,
-        inputs=[search_box],
-        outputs=[drug_dropdown, drug_display_to_id]
-    )
-    
-    drug_dropdown.change(
-        fn=store_selection,
-        inputs=[drug_dropdown, drug_display_to_id],
-        outputs=[selected_drug_id]
-    )
-    
-    analyze_btn.click(
-        fn=analyze_drug,
-        inputs=[
-            search_box,
-            selected_drug_id,
-            show_known_checkbox,
-            show_predicted_checkbox,
-            max_nodes_slider
-        ],
-        outputs=[
-            drug_info,
-            known_targets_table,
-            known_effects_table,
-            predicted_targets_table,
-            predicted_effects_table,
-            network_plot
-        ]
-    )
-
-# Launch
-if __name__ == "__main__":
-    demo.launch(share=False, server_name="0.0.0.0", server_port=7860)

cleanup.sh

@@ -1,59 +0,0 @@
-#!/bin/bash
-
-# Cleanup script to remove unnecessary files for end users
-# This keeps only essential files needed to run the Gradio app
-
-echo "🧹 Pharmacology Graph - Cleanup Script"
-echo "======================================"
-echo ""
-echo "This will remove development/duplicate files not needed to run the app."
-echo "Essential files (app.py, CSV data, requirements_app.txt) will be kept."
-echo ""
-read -p "Continue? (y/n) " -n 1 -r
-echo ""
-
-if [[ ! $REPLY =~ ^[Yy]$ ]]; then
-    echo "Cancelled."
-    exit 0
-fi
-
-echo ""
-echo "Removing unnecessary files..."
-
-# Documentation duplicates
-rm -f BUGFIX_SUMMARY.md && echo "  ✓ Removed BUGFIX_SUMMARY.md"
-rm -f APP_SUMMARY.md && echo "  ✓ Removed APP_SUMMARY.md"
-rm -f CHECKLIST.md && echo "  ✓ Removed CHECKLIST.md"
-
-# Duplicate/old scripts
-rm -f run_app.sh && echo "  ✓ Removed run_app.sh (use start_app.sh instead)"
-rm -f test_search.py && echo "  ✓ Removed test_search.py"
-
-# Old/temporary files (if they exist)
-rm -f main.py && echo "  ✓ Removed main.py"
-rm -f .python-version && echo "  ✓ Removed .python-version"
-
-# Image outputs (can regenerate from notebook)
-rm -f *.png && echo "  ✓ Removed PNG images"
-
-# Pickle files (can regenerate from notebook)
-rm -f *.pkl && echo "  ✓ Removed PKL files"
-
-# Intermediate CSV files (not needed for app)
-rm -f drug_mechanism_filtered.csv && echo "  ✓ Removed drug_mechanism_filtered.csv"
-rm -f drug_warnings.csv && echo "  ✓ Removed drug_warnings.csv"
-rm -f edges_drug_protein.csv && echo "  ✓ Removed edges_drug_protein.csv"
-rm -f proteins_for_embedding.csv && echo "  ✓ Removed proteins_for_embedding.csv"
-rm -f top_50_predicted_drug_targets.csv && echo "  ✓ Removed top_50_predicted_drug_targets.csv (duplicate)"
-
-echo ""
-echo "✅ Cleanup complete!"
-echo ""
-echo "Remaining essential files:"
-ls -lh *.py *.txt *.sh *.md 2>/dev/null | awk '{print "  - " $9 " (" $5 ")"}'
-echo ""
-echo "Data files:"
-ls -lh *.csv *.npy 2>/dev/null | awk '{print "  - " $9 " (" $5 ")"}'
-echo ""
-echo "Total size:"
-du -sh . | awk '{print "  " $1}'

graph_embeddings.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a5bfc5b74dcc2a159c88885b92432888d42a32101c34d7490051865be5d22dab
-size 2663040

index.html

@@ -0,0 +1,89 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8" />
+<meta name="viewport" content="width=device-width, initial-scale=1" />
+<title>Pharmacology Graph — Drug · Disease · Target Explorer</title>
+<link rel="icon" href="data:image/svg+xml,&lt;svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'&gt;&lt;text y='.9em' font-size='90'&gt;💊&lt;/text&gt;&lt;/svg&gt;" />
+<link rel="stylesheet" href="style.css?v=p5" />
+</head>
+<body>
+<header class="site-header">
+  <div class="wrap">
+    <h1>💊 Pharmacology Graph Explorer</h1>
+    <p class="tagline">
+      Search a <span class="chip drug">Drug</span> to see its predicted targets &amp; indications, or a
+      <span class="chip ind">Indication</span> / <span class="chip prot">Protein</span> to see the drugs most likely
+      linked to it — each flagged <span class="badge known">known</span> or <span class="badge novel">novel</span>.
+    </p>
+    <p class="meta-line" id="meta-line"></p>
+    <p class="meta-line" id="metrics-note"></p>
+  </div>
+</header>
+
+<main class="wrap">
+  <section class="search-box">
+    <label for="search" class="sr-only">Search</label>
+    <input id="search" type="text" autocomplete="off" spellcheck="false"
+           placeholder="Search a drug, disease/indication, or protein (e.g. sunitinib, breast neoplasms, CHEMBL231)…" />
+    <ul id="suggestions" class="suggestions" hidden></ul>
+  </section>
+
+  <section id="empty-state" class="empty">
+    <p>Start typing above. <span class="hint">Drugs · Indications · Proteins — all searchable, both directions.</span></p>
+    <div class="examples" id="examples"></div>
+  </section>
+
+  <section id="result" class="result" hidden>
+    <div class="drug-head">
+      <div>
+        <span class="chip" id="type-chip"></span>
+        <h2 id="drug-name"></h2>
+        <a id="drug-link" class="ext" target="_blank" rel="noopener"></a>
+      </div>
+      <div class="stat-cards" id="stat-cards"></div>
+    </div>
+
+    <div class="controls">
+      <div class="tabs" id="tabs" role="tablist"></div>
+      <div class="filters">
+        <div class="chips" id="status-filter">
+          <button class="chip active" data-status="all">All</button>
+          <button class="chip" data-status="novel">Novel only</button>
+          <button class="chip" data-status="known">Known only</button>
+        </div>
+        <input id="table-filter" class="table-filter" type="text" placeholder="Filter rows…" />
+        <div class="downloads">
+          <button id="dl-csv" class="dl">⬇ CSV</button>
+          <button id="dl-json" class="dl">⬇ JSON</button>
+        </div>
+      </div>
+    </div>
+
+    <table class="pred-table">
+      <thead>
+        <tr><th class="c-rank">#</th><th id="th-target">Target</th><th>ID</th><th class="c-score">Score</th><th class="c-status">Status</th></tr>
+      </thead>
+      <tbody id="pred-body"></tbody>
+    </table>
+    <p id="no-rows" class="no-rows" hidden>No rows match the current filter.</p>
+  </section>
+</main>
+
+<footer class="site-footer">
+  <div class="wrap">
+    <p>
+      Predictions are model outputs for research exploration only — not clinical or experimental validation.
+      Score is the model's relative confidence within each ranking (1.0 = top). Download full tables:
+      <a href="https://joevondahab.github.io/pharmacology-graph/data/predictions_drug_protein.csv">drug–protein CSV</a> ·
+      <a href="https://joevondahab.github.io/pharmacology-graph/data/predictions_drug_indication.csv">drug–indication CSV</a>.
+    </p>
+    <p class="src">Model: Ablation 2 “No Drug Graphs” (~3.29M params), faithfully reconstructed from
+      <a href="https://arxiv.org/abs/2603.01537" target="_blank" rel="noopener">arXiv:2603.01537</a> —
+      learnable drug embeddings + ESM-2 proteins + heterogeneous GraphSAGE message passing.</p>
+  </div>
+</footer>
+
+<script src="app.js?v=p5"></script>
+</body>
+</html>

node_to_idx.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:b88d5415499ea4b8732915f1459f14b330c79b377886ee2137b2d65f3131a15d
-size 55513

pyproject.toml

@@ -1,7 +0,0 @@
-[project]
-name = "pharmacology-graph"
-version = "0.1.0"
-description = "Add your description here"
-readme = "README.md"
-requires-python = ">=3.11"
-dependencies = []

requirements.txt

@@ -1,11 +0,0 @@
-# Hugging Face Spaces Requirements
-# Updated versions confirmed working (2025-10-11)
-
-gradio==4.44.1
-pandas==2.3.3
-numpy>=1.26.0,<2.0
-plotly==6.3.1
-networkx==3.2.1
-scikit-learn==1.4.0
-
-huggingface-hub==0.23.5

requirements_app.txt

@@ -1,14 +0,0 @@
-# Updated versions after testing (2025-01-24)
-# These versions are confirmed to work together without conflicts
-
-gradio==4.44.1
-pandas==2.3.3
-numpy>=1.26.0,<2.0  # NumPy 2.x breaks compatibility with pandas/numexpr
-plotly==6.3.1
-networkx==3.2.1
-scikit-learn==1.4.0==4.44.0
-pandas==2.2.0
-numpy==1.26.3
-plotly==5.18.0
-networkx==3.2.1
-scikit-learn==1.4.0

start_app.sh

@@ -1,24 +0,0 @@
-#!/bin/bash
-
-echo "🚀 Starting Pharmacology Graph App"
-echo "=================================="
-echo ""
-
-# Kill any existing instances
-pkill -f "python.*app.py" 2>/dev/null
-sleep 1
-
-# Clear port if needed
-lsof -ti:7860 | xargs kill -9 2>/dev/null
-sleep 1
-
-echo "✓ Port 7860 ready"
-echo "✓ Starting app..."
-echo ""
-echo "Open in browser: http://localhost:7860"
-echo "Press Ctrl+C to stop"
-echo ""
-
-# Start the app
-cd /home/joe/projects/pharmacology-graph
-python app.py

style.css

@@ -0,0 +1,98 @@
+:root{
+  --bg:#0f1420; --panel:#171e2e; --panel2:#1f2840; --line:#2b3550;
+  --text:#e7ecf5; --muted:#9aa7c2; --accent:#5b8cff; --accent2:#7c5cff;
+  --known:#3fb27f; --known-bg:rgba(63,178,127,.14);
+  --novel:#e0a83b; --novel-bg:rgba(224,168,59,.14);
+}
+*{box-sizing:border-box}
+body{margin:0;background:var(--bg);color:var(--text);
+  font:15px/1.5 system-ui,-apple-system,Segoe UI,Roboto,Helvetica,Arial,sans-serif}
+.wrap{max-width:960px;margin:0 auto;padding:0 20px}
+.sr-only{position:absolute;width:1px;height:1px;overflow:hidden;clip:rect(0 0 0 0)}
+a{color:var(--accent)}
+.site-header{background:linear-gradient(160deg,#1a2236,#0f1420);
+  border-bottom:1px solid var(--line);padding:34px 0 26px}
+h1{margin:0 0 6px;font-size:26px}
+.tagline{margin:0;color:var(--muted);max-width:640px}
+.tagline em{color:var(--text);font-style:normal;font-weight:600}
+.meta-line{margin:10px 0 0;color:var(--muted);font-size:13px}
+
+.search-box{position:relative;margin:26px 0 10px}
+#search{width:100%;padding:15px 18px;font-size:16px;border-radius:12px;
+  background:var(--panel);border:1px solid var(--line);color:var(--text);outline:none}
+#search:focus{border-color:var(--accent)}
+.suggestions{list-style:none;margin:6px 0 0;padding:6px;position:absolute;z-index:20;
+  width:100%;background:var(--panel2);border:1px solid var(--line);border-radius:12px;
+  max-height:320px;overflow:auto;box-shadow:0 18px 40px rgba(0,0,0,.45)}
+.suggestions li{padding:10px 12px;border-radius:8px;cursor:pointer;display:flex;
+  justify-content:space-between;gap:12px;align-items:center}
+.suggestions li.active,.suggestions li:hover{background:var(--panel)}
+.suggestions .sid{color:var(--muted);font-size:12px;font-variant-numeric:tabular-nums}
+.suggestions mark{background:transparent;color:var(--accent);font-weight:700}
+
+.empty{padding:40px 0;color:var(--muted);text-align:center}
+.empty .hint{display:block;margin-top:8px;font-size:13px;opacity:.8}
+.examples{display:flex;flex-wrap:wrap;gap:8px;justify-content:center;margin-top:20px}
+.examples button{background:var(--panel);border:1px solid var(--line);color:var(--text);
+  padding:7px 13px;border-radius:20px;cursor:pointer;font-size:13px}
+.examples button:hover{border-color:var(--accent)}
+
+.drug-head{display:flex;justify-content:space-between;gap:20px;flex-wrap:wrap;
+  align-items:flex-start;margin:8px 0 18px}
+#drug-name{margin:0;font-size:23px}
+.ext{font-size:13px}
+.stat-cards{display:flex;gap:10px;flex-wrap:wrap}
+.stat{background:var(--panel);border:1px solid var(--line);border-radius:10px;
+  padding:8px 14px;text-align:center;min-width:78px}
+.stat b{display:block;font-size:19px}
+.stat span{font-size:11px;color:var(--muted);text-transform:uppercase;letter-spacing:.04em}
+.stat.known b{color:var(--known)} .stat.novel b{color:var(--novel)}
+
+.controls{display:flex;justify-content:space-between;gap:14px;flex-wrap:wrap;
+  align-items:center;margin-bottom:12px}
+.tabs{display:flex;gap:4px;background:var(--panel);padding:4px;border-radius:10px;border:1px solid var(--line)}
+.tab{background:transparent;border:0;color:var(--muted);padding:8px 16px;border-radius:7px;cursor:pointer;font-size:14px}
+.tab.active{background:var(--accent);color:#fff}
+.filters{display:flex;gap:10px;align-items:center;flex-wrap:wrap}
+.chips{display:flex;gap:4px}
+.chip{background:var(--panel);border:1px solid var(--line);color:var(--muted);
+  padding:7px 12px;border-radius:7px;cursor:pointer;font-size:13px}
+.chip.active{color:var(--text);border-color:var(--accent)}
+.table-filter{background:var(--panel);border:1px solid var(--line);color:var(--text);
+  padding:8px 12px;border-radius:7px;outline:none;font-size:13px;width:140px}
+.downloads{display:flex;gap:6px}
+.dl{background:var(--panel2);border:1px solid var(--line);color:var(--text);
+  padding:8px 11px;border-radius:7px;cursor:pointer;font-size:13px}
+.dl:hover{border-color:var(--accent)}
+
+.pred-table{width:100%;border-collapse:collapse;font-size:14px}
+.pred-table th{text-align:left;color:var(--muted);font-weight:600;font-size:12px;
+  text-transform:uppercase;letter-spacing:.04em;padding:8px 10px;border-bottom:1px solid var(--line)}
+.pred-table td{padding:9px 10px;border-bottom:1px solid var(--panel2)}
+.pred-table tr:hover td{background:var(--panel)}
+.c-rank{width:38px;color:var(--muted)} .c-score{width:150px} .c-status{width:80px}
+.tname{font-weight:600}
+.tid a{font-size:12px;color:var(--muted)} .tid a:hover{color:var(--accent)}
+.scorewrap{display:flex;align-items:center;gap:8px}
+.bar{flex:1;height:7px;border-radius:5px;background:var(--panel2);overflow:hidden}
+.bar>i{display:block;height:100%;background:linear-gradient(90deg,var(--accent),var(--accent2))}
+.scoreval{font-variant-numeric:tabular-nums;color:var(--muted);font-size:12px;min-width:42px}
+.badge{font-size:11px;padding:3px 9px;border-radius:20px;font-weight:700;text-transform:uppercase;letter-spacing:.03em}
+.badge.known{color:var(--known);background:var(--known-bg)}
+.badge.novel{color:var(--novel);background:var(--novel-bg)}
+.no-rows{color:var(--muted);text-align:center;padding:24px}
+
+/* entity-type chips */
+.chip.drug,.chip.ind,.chip.prot{font-size:11px;padding:2px 8px;border-radius:20px;font-weight:700;
+  text-transform:uppercase;letter-spacing:.03em;border:0;cursor:default;vertical-align:middle}
+.chip.drug{color:#7fb2ff;background:rgba(91,140,255,.16)}
+.chip.ind{color:#e08bd0;background:rgba(224,139,208,.16)}
+.chip.prot{color:#6fd3c0;background:rgba(63,178,178,.16)}
+.tagline .chip{margin:0 1px}
+#type-chip{margin-bottom:6px;display:inline-block}
+.suggestions .s-left{display:flex;align-items:center;gap:8px;min-width:0}
+.suggestions .sname{overflow:hidden;text-overflow:ellipsis;white-space:nowrap}
+.examples button{display:inline-flex;align-items:center;gap:7px}
+
+.site-footer{margin-top:50px;border-top:1px solid var(--line);padding:22px 0 40px;color:var(--muted);font-size:13px}
+.site-footer .src{margin-top:6px;font-size:12px;opacity:.75}

test_app.py

@@ -1,157 +0,0 @@
-#!/usr/bin/env python3
-"""
-Quick test script to verify app.py data loading and functions work
-Run this before launching the full Gradio app to catch errors early
-"""
-
-import sys
-
-print("="*80)
-print("TESTING APP.PY DATA LOADING AND FUNCTIONS")
-print("="*80)
-
-# Test 1: Import libraries
-print("\n1. Testing library imports...")
-try:
-    import pandas as pd
-    import numpy as np
-    import networkx as nx
-    print("   ✓ Core libraries imported")
-except Exception as e:
-    print(f"   ✗ Error importing libraries: {e}")
-    sys.exit(1)
-
-try:
-    import gradio as gr
-    print("   ✓ Gradio imported")
-except Exception as e:
-    print(f"   ✗ Error importing Gradio: {e}")
-    print("   → Install with: uv pip install gradio")
-    sys.exit(1)
-
-try:
-    import plotly.graph_objects as go
-    print("   ✓ Plotly imported")
-except Exception as e:
-    print(f"   ✗ Error importing Plotly: {e}")
-    print("   → Install with: uv pip install plotly")
-    sys.exit(1)
-
-# Test 2: Load data files
-print("\n2. Testing data file loading...")
-
-required_files = {
-    'drug_nodes.csv': None,
-    'protein_nodes_with_embeddings.csv': None,
-    'drug_effects.csv': None,
-    'drugs_interactions.csv': None,
-    'graph_embeddings.npy': None,
-    'node_to_idx.npy': None
-}
-
-errors = []
-
-try:
-    drug_nodes = pd.read_csv('drug_nodes.csv')
-    required_files['drug_nodes.csv'] = f"✓ {len(drug_nodes)} drugs"
-    print(f"   ✓ drug_nodes.csv: {len(drug_nodes)} rows")
-except Exception as e:
-    errors.append(f"drug_nodes.csv: {e}")
-    print(f"   ✗ drug_nodes.csv: {e}")
-
-try:
-    protein_nodes = pd.read_csv('protein_nodes_with_embeddings.csv')
-    required_files['protein_nodes_with_embeddings.csv'] = f"✓ {len(protein_nodes)} proteins"
-    print(f"   ✓ protein_nodes_with_embeddings.csv: {len(protein_nodes)} rows")
-except Exception as e:
-    errors.append(f"protein_nodes_with_embeddings.csv: {e}")
-    print(f"   ✗ protein_nodes_with_embeddings.csv: {e}")
-
-try:
-    drug_effects = pd.read_csv('drug_effects.csv')
-    required_files['drug_effects.csv'] = f"✓ {len(drug_effects)} effects"
-    print(f"   ✓ drug_effects.csv: {len(drug_effects)} rows")
-except Exception as e:
-    errors.append(f"drug_effects.csv: {e}")
-    print(f"   ✗ drug_effects.csv: {e}")
-
-try:
-    drugs_interactions = pd.read_csv('drugs_interactions.csv')
-    required_files['drugs_interactions.csv'] = f"✓ {len(drugs_interactions)} interactions"
-    print(f"   ✓ drugs_interactions.csv: {len(drugs_interactions)} rows")
-except Exception as e:
-    errors.append(f"drugs_interactions.csv: {e}")
-    print(f"   ✗ drugs_interactions.csv: {e}")
-
-try:
-    embeddings = np.load('graph_embeddings.npy')
-    required_files['graph_embeddings.npy'] = f"✓ {embeddings.shape}"
-    print(f"   ✓ graph_embeddings.npy: shape {embeddings.shape}")
-except Exception as e:
-    errors.append(f"graph_embeddings.npy: {e}")
-    print(f"   ✗ graph_embeddings.npy: {e}")
-
-try:
-    node_to_idx = np.load('node_to_idx.npy', allow_pickle=True).item()
-    required_files['node_to_idx.npy'] = f"✓ {len(node_to_idx)} nodes"
-    print(f"   ✓ node_to_idx.npy: {len(node_to_idx)} nodes")
-except Exception as e:
-    errors.append(f"node_to_idx.npy: {e}")
-    print(f"   ✗ node_to_idx.npy: {e}")
-
-# Test 3: Check data structure
-if not errors:
-    print("\n3. Testing data structure...")
-    
-    # Check drug_nodes columns
-    required_cols = ['drug_internal_id', 'drug_id', 'drug_name', 'smile']
-    missing_cols = [col for col in required_cols if col not in drug_nodes.columns]
-    if missing_cols:
-        print(f"   ✗ drug_nodes missing columns: {missing_cols}")
-        errors.append(f"Missing columns in drug_nodes: {missing_cols}")
-    else:
-        print(f"   ✓ drug_nodes has all required columns")
-    
-    # Check protein_nodes columns
-    required_cols = ['protein_internal_id', 'protein_id', 'protein_name']
-    missing_cols = [col for col in required_cols if col not in protein_nodes.columns]
-    if missing_cols:
-        print(f"   ✗ protein_nodes missing columns: {missing_cols}")
-        errors.append(f"Missing columns in protein_nodes: {missing_cols}")
-    else:
-        print(f"   ✓ protein_nodes has all required columns")
-    
-    # Test search function
-    print("\n4. Testing search function...")
-    try:
-        # Search for common drug
-        test_queries = ['Morphine', 'Aspirin', 'CHEMBL25']
-        for query in test_queries:
-            matches = drug_nodes[
-                drug_nodes['drug_name'].str.lower().str.contains(query.lower(), na=False) |
-                drug_nodes['drug_id'].str.lower().str.contains(query.lower(), na=False)
-            ]
-            if len(matches) > 0:
-                print(f"   ✓ Search '{query}': found {len(matches)} results")
-            else:
-                print(f"   ⚠ Search '{query}': no results (might be OK)")
-    except Exception as e:
-        print(f"   ✗ Search function error: {e}")
-        errors.append(f"Search error: {e}")
-
-# Summary
-print("\n" + "="*80)
-if errors:
-    print("❌ TESTING FAILED")
-    print("\nErrors found:")
-    for i, error in enumerate(errors, 1):
-        print(f"  {i}. {error}")
-    print("\nPlease fix these errors before running the app.")
-    sys.exit(1)
-else:
-    print("✅ ALL TESTS PASSED!")
-    print("\nYour app should work correctly. Run it with:")
-    print("  ./run_app.sh")
-    print("  or")
-    print("  python app.py")
-    sys.exit(0)

top_50_predicted_drug_effects.csv

@@ -1,51 +0,0 @@
-drug_chembl_id,drug_name,effect_id,effect_name,similarity
-CHEMBL1257040,TRIFLUOPERAZINE HYDROCHLORIDE,D001714,Bipolar Disorder,0.545596
-CHEMBL198362,RIVAROXABAN,D009203,Myocardial Infarction,0.5439486
-CHEMBL435298,ESFLURBIPROFEN,D012216,Rheumatic Diseases,0.53732586
-CHEMBL1201782,FOSAPREPITANT DIMEGLUMINE,D009325,Nausea,0.5282381
-CHEMBL1709,SERTRALINE HYDROCHLORIDE,D016584,Panic Disorder,0.52576035
-CHEMBL1200470,ESOMEPRAZOLE SODIUM,D015043,Zollinger-Ellison Syndrome,0.5249922
-CHEMBL1951071,ARFORMOTEROL TARTRATE,D001986,Bronchial Spasm,0.51872176
-CHEMBL1744447,ROSUVASTATIN CALCIUM,D050171,Dyslipidemias,0.5174923
-CHEMBL2110372,RANITIDINE HYDROCHLORIDE,D004941,Esophagitis,0.5167754
-CHEMBL1138,EZETIMIBE,D050171,Dyslipidemias,0.5112109
-CHEMBL1201002,CARTEOLOL HYDROCHLORIDE,D009798,Ocular Hypertension,0.50021476
-CHEMBL3707246,OZANIMOD HYDROCHLORIDE,D003424,Crohn Disease,0.49325937
-CHEMBL4297088,CHLORHEXIDINE GLUCONATE,D003731,Dental Caries,0.49153596
-CHEMBL4297216,FEDRATINIB HYDROCHLORIDE,D013920,"Thrombocythemia, Essential",0.49133992
-CHEMBL1083385,CLOPIDOGREL BISULFATE,D011655,Pulmonary Embolism,0.49014294
-CHEMBL1027,TIAGABINE,D012640,Seizures,0.48950252
-CHEMBL2110372,RANITIDINE HYDROCHLORIDE,D005764,Gastroesophageal Reflux,0.48923716
-CHEMBL572,NITROFURANTOIN,D011704,Pyelonephritis,0.4890899
-CHEMBL503,LOVASTATIN,D050171,Dyslipidemias,0.48880434
-CHEMBL1059,PREGABALIN,D001007,Anxiety,0.4883793
-CHEMBL2105395,OSPEMIFENE,D016649,Primary Ovarian Insufficiency,0.48489913
-CHEMBL2105662,LOMITAPIDE MESYLATE,D050171,Dyslipidemias,0.48404413
-CHEMBL5314378,CASPOFUNGIN ACETATE,D016919,"Meningitis, Cryptococcal",0.48210338
-CHEMBL2218894,FLUVASTATIN SODIUM,D050171,Dyslipidemias,0.48126566
-CHEMBL3989844,DEXTROAMPHETAMINE SACCHARATE,D020181,"Sleep Apnea, Obstructive",0.48038968
-CHEMBL501,AMPHETAMINE SULFATE,D020181,"Sleep Apnea, Obstructive",0.4802664
-CHEMBL1650,CORTISONE ACETATE,D013167,"Spondylitis, Ankylosing",0.48004442
-CHEMBL1213250,CEFTOLOZANE SULFATE,D011704,Pyelonephritis,0.47739014
-CHEMBL1200402,AMLODIPINE BESYLATE,D050171,Dyslipidemias,0.47715414
-CHEMBL1364144,METHYLPREDNISOLONE ACETATE,D013167,"Spondylitis, Ankylosing",0.4765048
-CHEMBL406393,DEOXYCHOLIC ACID,D008105,"Liver Cirrhosis, Biliary",0.4760623
-CHEMBL3544971,DEXTROAMPHETAMINE SULFATE,D020181,"Sleep Apnea, Obstructive",0.4735379
-CHEMBL149972,PHENOBARBITAL SODIUM,D065768,Lennox Gastaut Syndrome,0.47247866
-CHEMBL1650,CORTISONE ACETATE,D006255,"Rhinitis, Allergic, Seasonal",0.46704072
-CHEMBL4650272,TRILACICLIB DIHYDROCHLORIDE,D018288,"Carcinoma, Small Cell",0.46695238
-CHEMBL1201782,FOSAPREPITANT DIMEGLUMINE,D000855,Anorexia,0.46403712
-CHEMBL3989923,PITAVASTATIN MAGNESIUM,D006938,Hyperlipoproteinemia Type II,0.46182984
-CHEMBL1650,CORTISONE ACETATE,D009205,Myocarditis,0.4610492
-CHEMBL1439,TAZOBACTAM SODIUM,D011704,Pyelonephritis,0.46001813
-CHEMBL1441059,ALBUTEROL SULFATE,D008173,"Lung Diseases, Obstructive",0.45689774
-CHEMBL650,METHYLPREDNISOLONE,D009879,"Ophthalmia, Sympathetic",0.45684493
-CHEMBL1200804,DICLOFENAC POTASSIUM,D013166,Spondylitis,0.45582262
-CHEMBL444633,RIFABUTIN,D001327,Autoimmune Diseases,0.45484185
-CHEMBL5314378,CASPOFUNGIN ACETATE,D009091,Mucormycosis,0.45365286
-CHEMBL137,METRONIDAZOLE,D012141,Respiratory Tract Infections,0.45224336
-CHEMBL675,ALENDRONATE SODIUM,D001847,Bone Diseases,0.45178664
-CHEMBL1269025,EDOXABAN,D054556,Venous Thromboembolism,0.4506775
-CHEMBL1200865,LOTEPREDNOL ETABONATE,D006255,"Rhinitis, Allergic, Seasonal",0.44942963
-CHEMBL1790041,RANITIDINE,D004381,Duodenal Ulcer,0.44910604
-CHEMBL89598,VIGABATRIN,D012640,Seizures,0.44849676

top_50_predicted_drug_protein.csv

@@ -1,51 +0,0 @@
-drug_chembl_id,drug_name,protein_id,protein_name,similarity
-CHEMBL3137320,TALAZOPARIB,CHEMBL2429709,Protein mono-ADP-ribosyltransferase PARP12,0.6001407
-CHEMBL3137320,TALAZOPARIB,CHEMBL2429708,Protein mono-ADP-ribosyltransferase PARP10,0.5721854
-CHEMBL1909072,PIPAMAZINE,CHEMBL1821,Muscarinic acetylcholine receptor M4,0.5399119
-CHEMBL419,MAFENIDE,CHEMBL3912,Carbonic anhydrase 13,0.5392196
-CHEMBL633,AMIODARONE,CHEMBL3371,5-hydroxytryptamine receptor 6,0.530897
-CHEMBL11,IMIPRAMINE,CHEMBL223,Alpha-1D adrenergic receptor,0.5215963
-CHEMBL17860,LOFEXIDINE,CHEMBL223,Alpha-1D adrenergic receptor,0.51608485
-CHEMBL1421,DASATINIB ANHYDROUS,CHEMBL2250,Tyrosine-protein kinase Blk,0.512127
-CHEMBL750,ZONISAMIDE,CHEMBL3510,Carbonic anhydrase 14,0.5120319
-CHEMBL1567,SUNITINIB MALATE,CHEMBL2007,Platelet-derived growth factor receptor alpha,0.5054646
-CHEMBL1173055,RUCAPARIB,CHEMBL2429709,Protein mono-ADP-ribosyltransferase PARP12,0.5037701
-CHEMBL5315125,DUVELISIB MONOHYDRATE,CHEMBL3267,"Phosphatidylinositol 4,5-bisphosphate 3-kinase catalytic subunit gamma isoform",0.5015378
-CHEMBL1963681,AVANAFIL,CHEMBL3977,"Cone cGMP-specific 3',5'-cyclic phosphodiesterase subunit alpha'",0.4979903
-CHEMBL1621,PALIPERIDONE,CHEMBL3371,5-hydroxytryptamine receptor 6,0.49721804
-CHEMBL762,OXYMETAZOLINE,CHEMBL217,D(2) dopamine receptor,0.4957183
-CHEMBL1078178,MOMELOTINIB,CHEMBL5970,Mitogen-activated protein kinase kinase kinase 3,0.49483612
-CHEMBL3301610,ABEMACICLIB,CHEMBL3529,Inhibitor of nuclear factor kappa-B kinase subunit epsilon,0.49270734
-CHEMBL3233142,LUMATEPERONE TOSYLATE,CHEMBL1899,5-hydroxytryptamine receptor 3A,0.49170905
-CHEMBL3301607,FILGOTINIB,CHEMBL1913,Platelet-derived growth factor receptor beta,0.48964554
-CHEMBL11,IMIPRAMINE,CHEMBL287,Sigma non-opioid intracellular receptor 1,0.48921353
-CHEMBL1535,HYDROXYCHLOROQUINE,CHEMBL245,Muscarinic acetylcholine receptor M3,0.48900294
-CHEMBL27,PROPRANOLOL,CHEMBL234,D(3) dopamine receptor,0.4874745
-CHEMBL4164059,PITOLISANT HYDROCHLORIDE,CHEMBL3371,5-hydroxytryptamine receptor 6,0.48707014
-CHEMBL669,CYCLOBENZAPRINE,CHEMBL1821,Muscarinic acetylcholine receptor M4,0.48594135
-CHEMBL978,METHACHOLINE,CHEMBL2035,Muscarinic acetylcholine receptor M5,0.48447487
-CHEMBL487253,BENDAMUSTINE,CHEMBL3310,Histone deacetylase 11,0.48373866
-CHEMBL86,METOCLOPRAMIDE,CHEMBL287,Sigma non-opioid intracellular receptor 1,0.48338604
-CHEMBL644,TRIMIPRAMINE,CHEMBL223,Alpha-1D adrenergic receptor,0.48287874
-CHEMBL808,ECONAZOLE,CHEMBL2035,Muscarinic acetylcholine receptor M5,0.4823854
-CHEMBL621,TRAZODONE,CHEMBL2056,D(1A) dopamine receptor,0.4819331
-CHEMBL1078178,MOMELOTINIB,CHEMBL3981,Serine/threonine-protein kinase 10,0.48185122
-CHEMBL726,FLUPHENAZINE,CHEMBL2035,Muscarinic acetylcholine receptor M5,0.48140538
-CHEMBL1421,DASATINIB ANHYDROUS,CHEMBL2007,Platelet-derived growth factor receptor alpha,0.48100707
-CHEMBL1123,DICYCLOMINE,CHEMBL222,Sodium-dependent noradrenaline transporter,0.4806236
-CHEMBL1834657,INFIGRATINIB PHOSPHATE,CHEMBL4722,Aurora kinase A,0.47917944
-CHEMBL1852688,INFIGRATINIB,CHEMBL5627,Leukocyte tyrosine kinase receptor,0.47882342
-CHEMBL671,THIOTEPA,CHEMBL1821,Muscarinic acetylcholine receptor M4,0.47861254
-CHEMBL24778,SILODOSIN,CHEMBL2056,D(1A) dopamine receptor,0.47838083
-CHEMBL3348923,TOVORAFENIB,CHEMBL2637,Mitogen-activated protein kinase 10,0.47791788
-CHEMBL81,RALOXIFENE,CHEMBL2035,Muscarinic acetylcholine receptor M5,0.47773397
-CHEMBL3301622,GILTERITINIB,CHEMBL3357,Interleukin-1 receptor-associated kinase 1,0.47708613
-CHEMBL1714574,TERCONAZOLE,CHEMBL231,Histamine H1 receptor,0.47602153
-CHEMBL644,TRIMIPRAMINE,CHEMBL1821,Muscarinic acetylcholine receptor M4,0.47474152
-CHEMBL1628227,DOXEPIN,CHEMBL233,Mu-type opioid receptor,0.47406977
-CHEMBL1633,KETOTIFEN FUMARATE,CHEMBL287,Sigma non-opioid intracellular receptor 1,0.47347492
-CHEMBL2035187,PACRITINIB,CHEMBL3778,Interleukin-1 receptor-associated kinase 4,0.47341335
-CHEMBL255863,NILOTINIB,CHEMBL3981,Serine/threonine-protein kinase 10,0.4726472
-CHEMBL1614701,SELUMETINIB,CHEMBL3938,Serine/threonine-protein kinase 16,0.47264248
-CHEMBL3545110,RIBOCICLIB,CHEMBL5261,Serine/threonine-protein kinase TAO1,0.47261065
-CHEMBL17860,LOFEXIDINE,CHEMBL2056,D(1A) dopamine receptor,0.47247177