README.md

metadata

license: mit
tags:
  - stxbp1
  - rare-disease
  - base-editing
  - genome-editing
  - crispr
  - epileptic-encephalopathy
  - translational-genomics
pretty_name: STXBP1 Base-Editing Parameter Sweep (Canonical 603-aa, v2)
task_categories:
  - tabular-classification
size_categories:
  - 100M<n<1B

STXBP1 Base-Editing Parameter Sweep — v2 (Canonical 603-aa frame)

TL;DR

170 pathogenic STXBP1 variants × 3,850,560 base-editing parameter combinations per variant = 654.6 million rows. Built on the MANE Plus Clinical canonical reference NM_003165.6 → NP_003156.1 (603 aa), the same frame ARIA and ClinVar use.

This is the canonical successor to SkyWhal3/STXBP1_Base_Editing_Parameter_Sweep (v1), which was built on NM_001032221.6 (594 aa MANE Select). For variants at protein positions 1–575, v1 and v2 results are byte-identical because the two transcripts encode identical residues in that region. v2 differs from v1 in two places:

Variant	v1 (594-aa frame)	v2 (603-aa canonical)
Q576X (c.1726C>T)	present but mislabeled — pos 576 is Q on 594-aa, T on canonical	removed (wrong-frame label; canonical[576]=T)
E603D (c.1809G>T)	errored (c.1809 past 594-aa CDS end at nt 1782)	present and correct

Reference frame

Field	Value
Transcript	NM_003165.6
Protein	NP_003156.1
UniProt	P61764-2 (MANE Plus Clinical)
Length	603 aa
CDS	1812 nt (including stop codon)
First 10 aa	MAPIGLKAVV
Last 4 aa	PTME
SHA-256 (protein)	f09d3d8cab1a46153031896bab6713a241b2eabd497cb2d354285d3a87756e6c

The canonical protein is pinned in stxbp1_canonical.py with a module-load SHA-256 assertion that fails loud if any byte of the sequence is modified.

Why two frames?

STXBP1 has two MANE-designated transcripts:

• MANE Select: NM_001032221.6 → P61764-1 (594 aa). Default in many older databases including UniProt's landing page.
• MANE Plus Clinical: NM_003165.6 → NP_003156.1 / P61764-2 (603 aa). What ClinVar reports variants against. What the clinical community uses. What ARIA and its Foundation partners use.

The two are byte-identical for residues 1–575 (which covers all common pathogenic variants like K196X, R292H, R406H) but diverge at the C-terminus. v2 exists because audiences evaluating ARIA and related tools expect variant-position data to be on the same frame as clinical reports.

What's in this dataset

Files

File	Size	Description
stxbp1_ULTIMATE_v2_canonical_best.json	~200 KB	Top-scoring parameter set per variant. Start here if you want one guide per variant.
stxbp1_ULTIMATE_v2_canonical_sweep.csv.gz	~5 GB	All 654.6 M param combos, gzipped. Full search space for your own re-ranking.
stxbp1_snv_variants_170_canonical.txt	8 KB	The 170-variant input list with provenance comments.

Schema (CSV)

Column	Type	Description
label	str	Short variant label (e.g. K196X)
cdna	str	c.DNA change (e.g. c.586A>T) — on NM_003165.6
edit_type	str	ABE (A>G, T>C), CBE (C>T, G>A), or PRIME
scan	int	Guide search radius in nt (10–200)
ctx	int	Context identity window in nt (5–100)
win	str	Edit-window range (e.g. 4-8)
identity	float	Human/mouse sequence identity in the context window (0–1)
candidates	int	Number of PAM-containing guides found in scan window
best_mm	int	Mismatches between best human guide and mouse at same site
in_window	bool	Whether the edit site falls inside the edit window
compat	bool	Pass/fail composite: identity ≥ 0.90 AND mm ≤ 3 AND in_window
score	float	0–100 composite score (identity × 40 + guide quality)

The CSV header includes three comment rows (#-prefixed) with the reference frame, build date, and variant list filename.

Quick-start

import pandas as pd
best = pd.read_json("stxbp1_ULTIMATE_v2_canonical_best.json", orient="index")
best.loc[best.index.str.startswith("K196X")]

# Or stream the full sweep for one variant
import pandas as pd
iter = pd.read_csv("stxbp1_ULTIMATE_v2_canonical_sweep.csv.gz", chunksize=1_000_000)
for chunk in iter:
    k196x = chunk[chunk["label"] == "K196X"]
    if not k196x.empty:
        print(k196x.sort_values("score", ascending=False).head())

Reproducibility

Pipeline source: run_canonical_parameter_sweep.py with the canonical-input file committed alongside. The build script:

1. Fetches NM_003165.6 + NM_011502 CDS live from NCBI (cached on first run).
2. Reads the 170-variant canonical input file.
3. Runs one variant at a time through a CPU multiprocessing pool over the 191 × 96 × 210 = 3,850,560 parameter grid.
4. Writes the CSV with a banner header declaring the reference frame.

The canonical assertion test (assert_canonical_state.py) should be run before each rebuild to confirm the reference is still pinned at the SHA-256 above.

Provenance and prior work

• v1 (SkyWhal3/STXBP1_Base_Editing_Parameter_Sweep, Jan 2026): built on NM_001032221.6 (594 aa). Archived-not-deprecated — valid for variants 1–575. This v2 supersedes it for demo-facing and publication-facing use.
• Session 96 (2026-04-24): a wider canonical rebuild corrected a years-old contamination in an adjacent ARIA dataset that had been folded against a 586-aa non-Munc18-1 protein. The 603-aa frame lock-in described here is the downstream of that fix.

Citation

If you use this dataset, please cite both the v1 build (which produced the original 170-variant input list and scoring methodology) and this v2 rebuild for the canonical frame correction.

License

MIT (matches v1).

Contact

Adam Freygang — STXBP1 parent, ARIA developer.