Block-wise ablation, manuscript preview tooling, repaired binding sites - #4
Open
hallelx2 wants to merge 1 commit into
Open
Block-wise ablation, manuscript preview tooling, repaired binding sites#4hallelx2 wants to merge 1 commit into
hallelx2 wants to merge 1 commit into
Conversation
…ding sites Separate the docking arm into blocks that carry different information rather than ablating it as one lump. Eleven columns describe the docked *ligand's* 3D shape and need no receptor to compute; 132 are receptor-dependent. The combined ablation could not distinguish them, so it could not answer whether receptor structure helps. Measured over three seeds x three folds under stratified, scaffold-grouped and receptor-grouped CV, the receptor-dependent block contributes -0.002 / +0.000 / +0.004 macro-F1 -- each inside the seed spread of its own baseline. Also: - build_preview_html.py renders a reading preview with figures placed at first mention; the DOCX pipeline puts them under the legends, which makes a draft hard to read. - binding_sites.json carries the repaired boxes for the seven receptors whose written chain was a G-protein subunit, a nanobody or a different receptor, plus a repaired_receptors record. main still shipped the stale boxes. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The previous ablation dropped the whole docking arm as one block. That block mixed three different information sources, so it could not answer whether receptor structure helps — 11 of its columns describe the docked ligand's own 3D shape and need no receptor at all.
What changed
scripts/ablation_by_block.py— ablates each block separately under stratified, scaffold-grouped and receptor-grouped CV (3 seeds × 3 folds). Contribution = full − without:Every value is smaller than the seed spread of its own baseline, so none is distinguishable from zero. The receptor-dependent block contributes nothing on the split that tests it.
scripts/build_preview_html.py— renders a reading preview with each figure placed at its first mention, rather than under the legends at the end.data/processed/binding_sites.json— repaired boxes for the seven receptors whose written chain was a G-protein subunit, a nanobody, or a different receptor entirely, plus arepaired_receptorsrecord.mainstill shipped the stale boxes.Note
The earlier ablation ran
StratifiedKFoldwhile being reported as scaffold-grouped. The delta itself was right; the split label was not. Figures now carry split, seeds, folds, baseline spread and sign convention explicitly.🤖 Generated with Claude Code