Document f1 and f2 checkpoint sets

expression_models/README.md

@@ -5,17 +5,19 @@ EPInformer pipeline.
 
 ## Gene-expression checkpoints
 
-The `expression_models/` directory contains the validated `f3` models with
-three enhancer features (distance, activity, and Hi-C contact) and promoter
-signal:
+The `expression_models/` directory contains one checkpoint for each of the
+12 leave-chromosome-out folds. `f1` uses distance only, `f2` uses distance and
+activity, and `f3` uses distance, activity, Hi-C contact, and promoter signal.
+The `f3` files are directly in each assay directory; `f1` and `f2` are in the
+corresponding subdirectories.
 
-| Cell type | Assay | Directory | Pooled Pearson R |
-|---|---|---|---:|
-| K562 | RNA | `expression_models/K562/RNA` | 0.8564 |
-| K562 | CAGE | `expression_models/K562/CAGE` | 0.8673 |
-| GM12878 | RNA | `expression_models/GM12878/RNA` | 0.8602 |
-| GM12878 | CAGE | `expression_models/GM12878/CAGE` | 0.8902 |
+| Cell type | Assay | f1 | f2 | f3 |
+|---|---|---:|---:|---:|
+| K562 | RNA | 0.7528 | 0.8552 | 0.8564 |
+| K562 | CAGE | 0.8114 | 0.8709 | 0.8673 |
+| GM12878 | RNA | 0.7610 | 0.8268 | 0.8602 |
+| GM12878 | CAGE | 0.8270 | 0.8687 | 0.8902 |
 
-Each directory contains one checkpoint for each fold (1--12). Use the
-matching encoder checkpoints, HDF5 input, expression labels, and fold split
-described in the [EPInformer pipeline README](https://github.com/pinellolab/EPInformer/tree/pipeline).
+Values are pooled Pearson R. Use the matching encoder checkpoints, HDF5 input,
+expression labels, and fold split described in the
+[EPInformer pipeline README](https://github.com/pinellolab/EPInformer/tree/pipeline).