Fix #575

alphapulldown/scripts/convert_to_modelcif.py

@@ -17,7 +17,7 @@ import zipfile
 import glob
 import ast
 
-from Bio.PDB import PDBParser, PPBuilder
+from Bio.PDB import PDBParser, MMCIFParser, PPBuilder
 from Bio.PDB.Structure import Structure as BioStructure
 from absl import app, flags, logging
 import numpy as np
@@ -792,6 +792,7 @@ def _get_feature_metadata(
     modelcif_json: dict,
     cmplx_name: str,
     out_dir: list,
+    fallback_structure_path: str | None = None,
 ) -> Tuple[List[str], List[str]]:
     """Read metadata from a feature JSON file."""
     if "__meta__" not in modelcif_json:
@@ -813,12 +814,49 @@ def _get_feature_metadata(
                 modelcif_json["__meta__"][mnmr][
                     "software"
                 ] = _get_software_with_parameters(jdata["software"], jdata["other"])
-                fp = jdata["other"]["fasta_paths"]
-                fp = ast.literal_eval(fp)
-                for curr_seq, curr_desc in iter_seqs(fp):
-                    new_entry = {'description': curr_desc, 'sequence': curr_seq}
-                    if new_entry not in fasta_dicts:
-                        fasta_dicts.append(new_entry)
+                fp = jdata["other"].get("fasta_paths")
+                if fp is not None:
+                    fp = ast.literal_eval(fp)
+                    existing_fp = []
+                    missing_fp = []
+                    for p in fp:
+                        if os.path.isfile(p):
+                            existing_fp.append(p)
+                        else:
+                            # Also try relative-to-output-dir paths, as users
+                            # often run the converter from a different cwd.
+                            relp = os.path.join(out_dir, p)
+                            if os.path.isfile(relp):
+                                existing_fp.append(relp)
+                            else:
+                                missing_fp.append(p)
+                    if missing_fp:
+                        logging.warning(
+                            "FASTA file(s) referenced in feature metadata not found: "
+                            + ", ".join(missing_fp)
+                        )
+                    for curr_seq, curr_desc in iter_seqs(existing_fp):
+                        new_entry = {"description": curr_desc, "sequence": curr_seq}
+                        if new_entry not in fasta_dicts:
+                            fasta_dicts.append(new_entry)
+
+    # Fallback: if FASTA input is missing, we still want a usable sequence
+    # mapping (primarily for descriptions) by extracting polymer sequences from
+    # the model structure file (PDB or mmCIF).
+    if not fasta_dicts and fallback_structure_path:
+        try:
+            fasta_dicts = _get_fasta_dicts_from_structure_file(
+                fallback_structure_path, cmplx_name
+            )
+            logging.warning(
+                "No FASTA sequences available; extracted sequences from structure file "
+                + f"'{fallback_structure_path}'."
+            )
+        except Exception as exc:
+            logging.warning(
+                "No FASTA sequences available and failed to extract sequences from "
+                + f"'{fallback_structure_path}': {exc}"
+            )
 
     return cmplx_name, fasta_dicts
 
@@ -852,8 +890,8 @@ def _get_entities(
         # Using MD5 sums for comparing sequences
         sequences[hashlib.md5(seq.encode()).hexdigest()] = description
 
-    # gather molecular entities from PDB file
-    structure = PDBParser().get_structure(cmplx_name, pdb_file)
+    # gather molecular entities from structure file (PDB or mmCIF)
+    structure = _load_structure(cmplx_name, pdb_file)
     cif_json["target_entities"] = []
     already_seen = []
     for seq in PPBuilder(radius=999999999).build_peptides(structure, aa_only=False):
@@ -880,6 +918,41 @@ def _get_entities(
     return structure
 
 
+def _load_structure(structure_id: str, structure_path: str) -> BioStructure:
+    """Load a structure from PDB or mmCIF path."""
+    lower = structure_path.lower()
+    if lower.endswith(".cif") or lower.endswith(".mmcif"):
+        # QUIET to suppress warnings on common mmCIF oddities
+        return MMCIFParser(QUIET=True).get_structure(structure_id, structure_path)
+    return PDBParser(QUIET=True).get_structure(structure_id, structure_path)
+
+
+def _get_fasta_dicts_from_structure_file(
+    structure_path: str, structure_id: str
+) -> List[dict]:
+    """Extract polymer sequences from a PDB/mmCIF structure file.
+
+    Returns a list of dicts with keys: 'description', 'sequence'.
+    """
+    structure = _load_structure(structure_id, structure_path)
+    fasta_dicts: List[dict] = []
+    seen = set()
+    polypeptides = PPBuilder(radius=999999999).build_peptides(structure, aa_only=False)
+    for pp in polypeptides:
+        if not pp:
+            continue
+        chain_id = pp[0].parent.id
+        seq = str(pp.get_sequence())
+        if not seq:
+            continue
+        # De-duplicate identical sequences across chains
+        if seq in seen:
+            continue
+        seen.add(seq)
+        fasta_dicts.append({"description": f"chain_{chain_id}", "sequence": seq})
+    return fasta_dicts
+
+
 def _get_scores(cif_json: dict, scr_file: str) -> None:
     """Add scores to JSON data."""
     # Read from jsons instead
@@ -1233,7 +1306,7 @@ def alphapulldown_model_to_modelcif(
     modelcif_json = {}
     # fetch metadata
     cmplx_name, fasta_dicts = _get_feature_metadata(
-        modelcif_json, cmplx_name, out_dir
+        modelcif_json, cmplx_name, out_dir, fallback_structure_path=mdl[0]
     )
     # fetch/ assemble more data about the modelling experiment
     _get_model_info(

test/test_modelcif.py

@@ -6,6 +6,8 @@ import shutil
 import tempfile
 from os.path import join, dirname, abspath
 import zipfile
+import json
+import glob
 
 """
 Test conversion of PDB to CIF for monomers and multimers
@@ -119,3 +121,34 @@ class TestConvertPDB2CIF(parameterized.TestCase):
             command.extend(["--model_selected", str(model_selected)])
 
         return command
+
+    def test_missing_fasta_falls_back_to_structure_sequence(self):
+        """If FASTA path in feature metadata is missing, parse sequence from structure."""
+        with tempfile.TemporaryDirectory() as temp_dir:
+            test_output_dir = join(temp_dir, "output")
+            shutil.copytree(join(self.input_dir, "TEST"), test_output_dir)
+
+            # Break the FASTA reference in feature metadata.
+            md_files = glob.glob(join(test_output_dir, "*_feature_metadata_*.json"))
+            self.assertTrue(md_files, "No feature metadata JSON found in test output dir")
+            for md_file in md_files:
+                with open(md_file, "r", encoding="ascii") as fh:
+                    data = json.load(fh)
+                data["other"]["fasta_paths"] = "['/this/path/does/not/exist.fasta']"
+                with open(md_file, "w", encoding="ascii") as fh:
+                    json.dump(data, fh, indent=2)
+
+            command = self.build_command(
+                test_output_dir, add_associated=False, compress=False, model_selected=0
+            )
+            subprocess.run(command, check=True, capture_output=True, text=True)
+
+            out_cif = join(test_output_dir, "ranked_0.cif")
+            self.assertTrue(os.path.exists(out_cif), "ModelCIF output was not created")
+
+            # Sequence should still be present in the output even without FASTA.
+            # The TEST sequence starts with "MESAIA..." in test FASTA and in the
+            # structure-derived sequence.
+            with open(out_cif, "r", encoding="ascii") as fh:
+                cif_txt = fh.read()
+            self.assertIn("MESAIA", cif_txt)