Notebooks

Notebooks

Three notebooks ship with the reproducible_pipeline branch (figures are pre-embedded, so results are visible without re-running):

predict_enhancer_and_expression.ipynb

The main demo, covering both models:

• Part 1 — a faithful, cell-for-cell reproduction of upstream's predict_enhancer_activity.ipynb: predicts K562 enhancer activity across the CRISPRi-validated KLF1 enhancer and runs in-silico saturation mutagenesis (ISM) to expose the TF motif, using the published fold-8 encoder.
• Part 2 — live 12-fold cross-chromosome inference for K562 RNA (R ≈ 0.856) and CAGE (R ≈ 0.867), plus the cross-cell RNA table.

reproduce_github_enhancer_activity.ipynb

A standalone, upstream-exact reproduction of predict_enhancer_activity.ipynb (KLF1 activity barplot + ISM logo). Uses the Fulco-2019 CRISPRi-FlowFISH table and the hg19 genome, matching the original notebook cell-for-cell.

visualize_results.ipynb

Lightweight evaluation dashboards built from the saved per-fold prediction CSVs: density scatter, per-fold Pearson R, and the cross-cell results table.

Checkpoints

The demos load the pretrained encoder from JiecongLin/EPInformer-reproducible (or trained_models/ on the branch). The KLF1 demo uses enhancer_encoders/K562/fold_8.pt.