How d7125/repression-domain-transfer-audit works
This repository is a reproducible analysis project auditing whether sequence models trained on human repression-domain screens transfer from the rTetR assay to the dCas9/CD43 context. Its main workflow assembles screen-derived tile data, creates homology-aware splits and ESM-2 embeddings, builds a shared master table, and evaluates regression, classification, and transfer. Additional branches test retest validity, motif saliency, composition and domain annotations, variance, and robustness, with figure generation for reporting. The README documents the data-preparation order; unsampled analysis links are omitted where call relationships are not evidenced.
Main entry points: build_master.py Shared tile table · cnn_model.py Motif CNN models · train_cnn.py Training and splits · cd43_topk2.py CD43 transfer ranking
Components 25
Data preparation
- Tile preparation
aduy_code/predata_v2.py - Tile table preparation
aduy_code/prepare_data.py - Protein FASTA
aduy_code/make_fasta.py - Homology clustering
aduy_code/cluster_domains.py - ESM-2 embeddings
aduy_code/extract_per_residue.py - Shared tile table
aduy_code/build_master.py
Sequence models
- Motif CNN models
aduy_code/cnn_model.py - Training and splits
aduy_code/train_cnn.py - Classification training
aduy_code/train_classify.py
Transfer evaluation
- CD43 regression
aduy_code/cd43_model.py - Retest validation
aduy_code/retest_validate.py - Retest seed analysis
run/rev2/retest_seeds.py - CD43 transfer ranking
run/rev2/cd43_topk2.py - Measured-feature baseline
run/rev2/cd43_tabular_baseline.py - CD43 robustness variants
run/rev3/cd43_variants.py - Confirmed-hit classifier
aduy_code/classify_confirmed.py
Biological analyses
- Motif saliency
aduy_code/saliency_vs_motif.py - Saliency controls
run/rev2/saliency_seeds.py - Composition controls
aduy_code/composition_control.py - Domain annotations
aduy_code/pfam_annotate.py - Variance decomposition
aduy_code/icc_variance.py - Robustness analyses
run/rev3
Results and figures
- Paper figures
figures/make_paper_figures.py - Top-k figures
figures/make_fig_topk.py
Other components
- Published screen tables
Connections 39
- Published screen tables → Tile preparation: provides tables
README.md - Tile preparation → Tile table preparation: prepares tiles
README.md - Tile table preparation → Protein FASTA: feeds sequences
README.md - Protein FASTA → Homology clustering: clusters proteins
README.md - Homology clustering → Training and splits: groups splits
README.md - Tile table preparation → ESM-2 embeddings: supplies sequences
README.md - ESM-2 embeddings → Shared tile table: precedes build
README.md - Tile table preparation → Shared tile table: builds shared table
README.md - Motif CNN models → Training and splits: defines model
aduy_code/train_cnn.py - Shared tile table → CD43 regression: provides measurements
aduy_code/cd43_model.py - Motif CNN models → CD43 regression: provides model
aduy_code/cd43_model.py - Training and splits → CD43 regression: provides training
aduy_code/cd43_model.py - Shared tile table → Retest validation: provides measurements
aduy_code/retest_validate.py - Motif CNN models → Retest validation: provides model
aduy_code/retest_validate.py - Training and splits → Retest validation: provides training
aduy_code/retest_validate.py - Shared tile table → Retest seed analysis: provides measurements
run/rev2/retest_seeds.py - Training and splits → Retest seed analysis: provides training
run/rev2/retest_seeds.py - Shared tile table → CD43 transfer ranking: provides measurements
run/rev2/cd43_topk2.py - Motif CNN models → CD43 transfer ranking: provides model
run/rev2/cd43_topk2.py - Training and splits → CD43 transfer ranking: provides training
run/rev2/cd43_topk2.py - Shared tile table → Measured-feature baseline: provides features
run/rev2/cd43_tabular_baseline.py - CD43 transfer ranking → Measured-feature baseline: reads predictions
run/rev2/cd43_tabular_baseline.py - Shared tile table → CD43 robustness variants: provides measurements
run/rev3/cd43_variants.py - Motif CNN models → CD43 robustness variants: provides model
run/rev3/cd43_variants.py - Training and splits → CD43 robustness variants: provides training
run/rev3/cd43_variants.py - Shared tile table → Confirmed-hit classifier: provides labels
aduy_code/classify_confirmed.py - Motif CNN models → Confirmed-hit classifier: provides model
aduy_code/classify_confirmed.py - Classification training → Confirmed-hit classifier: runs classifier
aduy_code/classify_confirmed.py - Shared tile table → Motif saliency: provides sequences
aduy_code/saliency_vs_motif.py - Motif CNN models → Motif saliency: provides model
aduy_code/saliency_vs_motif.py - Training and splits → Motif saliency: provides training
aduy_code/saliency_vs_motif.py - Shared tile table → Saliency controls: provides measurements
run/rev2/saliency_seeds.py - Motif CNN models → Saliency controls: provides model
run/rev2/saliency_seeds.py - Shared tile table → Variance decomposition: provides measurements
README.md - Shared tile table → Composition controls: provides measurements
README.md - Shared tile table → Domain annotations: provides sequences
- Shared tile table → Robustness analyses: supports variants
README.md - CD43 transfer ranking → Paper figures: reports results
- CD43 transfer ranking → Top-k figures: reports ranking