fix three bugs in dynamics/analysis.py (dropped hinges, IndexError, wrong collectivity norm) - #2262
Open
steps-re wants to merge 1 commit into
Open
fix three bugs in dynamics/analysis.py (dropped hinges, IndexError, wrong collectivity norm)#2262steps-re wants to merge 1 commit into
steps-re wants to merge 1 commit into
Conversation
…rong collectivity norm getGlobalHinges was silently discarding hinges from every chain but the last: the accumulation block was indented one level too high, outside the per-chain loop, so it only ever ran once using the leftover h/l/fst from the final chain. Moved it inside the loop. Also fixed an off-by-one in the same function: chains stores inclusive (fst, lst) pairs but the vecs slice used the exclusive vecs[fst:lst], dropping the last residue of every chain from hinge detection. Now vecs[fst:lst+1, i]. getHinges raised IndexError (regs[0] on an empty list) whenever an eigenvector segment had no sign change at all. Added a guard that returns an empty hinge list in that case, matching the function's normal return type. calcCollectivity normalized u2in by dividing by sqrt(sum) instead of sum when masses were supplied, which doesn't produce a valid probability distribution per Bruschweiler 1995 eq. 5 (the exponent no longer sums the correct entropy term). Only the masses path was affected; the default unit-mass path is unchanged since sum and sqrt(sum) both equal 1 there. Added prody/tests/dynamics/test_analysis_hinge_bugs.py with synthetic, offline, deterministic repros for all three (multi-chain AtomGroup + monkeypatched _getModeVecs for the hinge bugs, hand-computed Bruschweiler values for collectivity). All 6 new tests fail on the prior code and pass after the fix; existing dynamics test modules show no regressions. Co-Authored-By: Claude <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.
disclosing up front: this was put together with help from claude (anthropic's coding agent), then reproduced and verified by me before opening. first pr i'm opening on prody.
three separate bugs in
prody/dynamics/analysis.py, all found while looking at the hinge-finding code. each is small and independent so happy to split into separate prs if you'd rather.1. getGlobalHinges silently drops every chain's hinges but the last
the
if trim is not False / elseblock that accumulates hinges is indented at the same level asfor fst, lst in chains:, so it runs once after the chain loop finishes, using only the leftoverh/l/fstfrom the last chain. for any multi-chain structure, every earlier chain's hinges are silently thrown away. fix is to move the accumulation inside the loop.there's also an off-by-one in the same spot:
chainsstores inclusive(fst, lst)pairs, but the code slicesvecs[fst:lst, i](exclusive), which drops each chain's last residue from hinge detection. fixed tovecs[fst:lst+1, i].2. getHinges crashes on a segment with no sign change
merged = [regs[0]]raises IndexError whenregsis empty, which happens whenever an eigenvector segment has no sign crossing (realistic for short segments, and more likely once bug #1 is splitting by chain). added a guard to return no hinges instead of crashing.3. calcCollectivity mis-normalizes when masses are supplied
Bruschweiler 1995 needs the per-atom msd normalized into a probability distribution (
p_i = u_i^2 / sum(u_j^2)) before the shannon-entropy step. the code divides bysqrt(sum(u_j^2))instead. this cancels on the default path because mode vectors are already unit-normalized (sum == sqrt(sum) == 1), so it only shows up whenmassesis passed, which is the documented-supported path. for v=[0.6,0.8], masses=[1,4] the current result is ~0.9878 vs the correct ~0.9269 (~35% off). fix is to divide by the sum, not its sqrt.verification
added prody/tests/dynamics/test_analysis_hinge_bugs.py (offline, deterministic, no downloads): 4 tests targeting the bugs + 2 control/regression guards. confirmed all 4 fail on current main (IndexError, dropped-chain hinge mismatch, wrong collectivity value) and pass after the fixes. ran the full prody/tests/dynamics/ suite, no regressions (existing test_hinge_finder.py still passes). the default no-masses collectivity path is unchanged.
mike