From 8356d810a83359111fd0c994cbe07b8ca1c60a4c Mon Sep 17 00:00:00 2001
From: Xavier Robin <xavier.robin@unibas.ch>
Date: Fri, 26 Apr 2024 13:35:28 +0200
Subject: [PATCH] ligand scoring: fix infinite loop in assignment
---
modules/mol/alg/pymod/ligand_scoring.py | 21 +++++++++-----
modules/mol/alg/tests/test_ligand_scoring.py | 29 ++++++++++++++++++++
2 files changed, 43 insertions(+), 7 deletions(-)
diff --git a/modules/mol/alg/pymod/ligand_scoring.py b/modules/mol/alg/pymod/ligand_scoring.py
index 4c7ebdf54..ce98690dd 100644
--- a/modules/mol/alg/pymod/ligand_scoring.py
+++ b/modules/mol/alg/pymod/ligand_scoring.py
@@ -695,11 +695,6 @@ class LigandScorer:
# Disconnected graph is handled elsewhere
continue
- substructure_match = len(symmetries[0][0]) != len(model_ligand.atoms)
- coverage = len(symmetries[0][0]) / len(model_ligand.atoms)
- self._assignment_match_coverage[target_id, model_id] = coverage
- self._assignment_isomorphisms[target_id, model_id] = 1.
-
################################################################
# Compute best rmsd/lddt-pli by naively enumerating symmetries #
################################################################
@@ -713,13 +708,25 @@ class LigandScorer:
###########################################
# Extend results by symmetry related info #
###########################################
+ if (rmsd_result is None) != (lddt_pli_result is None):
+ # Ligand assignment makes assumptions here, and is likely
+ # to not work properly if this differs. There is no reason
+ # it would ever do, so let's just check it
+ raise Exception("Ligand scoring bug: discrepency between "
+ "RMSD and lDDT-PLI definition.")
if rmsd_result is not None:
+ # Now we assume both rmsd_result and lddt_pli_result are defined
+ # Add coverage
+ substructure_match = len(symmetries[0][0]) != len(model_ligand.atoms)
+ coverage = len(symmetries[0][0]) / len(model_ligand.atoms)
+ self._assignment_match_coverage[target_id, model_id] = coverage
+ self._assignment_isomorphisms[target_id, model_id] = 1.
+ # Add RMSD
rmsd_result["substructure_match"] = substructure_match
rmsd_result["coverage"] = coverage
if self.unassigned:
rmsd_result["unassigned"] = False
-
- if lddt_pli_result is not None:
+ # Add lDDT-PLI
lddt_pli_result["substructure_match"] = substructure_match
lddt_pli_result["coverage"] = coverage
if self.unassigned:
diff --git a/modules/mol/alg/tests/test_ligand_scoring.py b/modules/mol/alg/tests/test_ligand_scoring.py
index 0eb503787..6755b0e8e 100644
--- a/modules/mol/alg/tests/test_ligand_scoring.py
+++ b/modules/mol/alg/tests/test_ligand_scoring.py
@@ -731,6 +731,35 @@ class TestLigandScoring(unittest.TestCase):
sc.unassigned_target_ligands["F"][1] == "symmetries"
+ def test_no_binding_site(self):
+ """
+ Test the behavior when there's no binding site in proximity of
+ the ligand. This test was introduced to identify some subtle issues
+ with the ligand assignment that can cause it to enter an infinite
+ loop when the data matrices are not filled properly.
+ """
+ trg = _LoadMMCIF("1r8q.cif.gz").Copy()
+ mdl = trg.Copy()
+
+ trg_zn = trg.FindResidue("H", 1)
+ trg_g3d = trg.FindResidue("F", 1)
+
+
+ # Move the zinc out of the reference binding site...
+ ed = trg.EditXCS()
+ ed.SetAtomPos(trg_zn.FindAtom("ZN"),
+ trg_zn.FindAtom("ZN").pos + geom.Vec3(6, 0, 0))
+ # Remove some atoms from G3D to decrease coverage. This messed up
+ # the assignment in the past.
+ ed.DeleteAtom(trg_g3d.FindAtom("O6"))
+ ed.UpdateICS()
+
+ sc = LigandScorer(mdl, trg, target_ligands=[trg_zn, trg_g3d],
+ coverage_delta=0, substructure_match=True)
+ self.assertTrue(np.isnan(sc.rmsd_matrix[0, 3]))
+ self.assertEqual(sc.unassigned_target_ligands["H"][1], "binding_site")
+
+
def test_no_lddt_pli_contact(self):
"""
Test behaviour where a binding site has no lDDT-PLI contacts.
--
GitLab