Add BioPandas to facilitate recovery of any missing PDB chain IDs dur…

…ing inference

project/utils/deepinteract_utils.py

@@ -1,3 +1,4 @@
+import difflib
 import itertools
 import logging
 import os
@@ -27,6 +28,7 @@
 from Bio.Align import MultipleSeqAlignment
 from Bio.Seq import Seq
 from Bio.SeqRecord import SeqRecord
+from biopandas.pdb import PandasPdb
 from pytorch_lightning.loggers import TensorBoardLogger, WandbLogger
 
 from project.utils.deepinteract_constants import FEAT_COLS, ALLOWABLE_FEATS, D3TO1
@@ -593,9 +595,11 @@ def copy_input_to_raw_dir(input_dataset_dir: str, pdb_filepath: str, pdb_code: s
     """Make a copy of the input PDB file in the newly-created raw directory."""
     filename = db.get_pdb_code(pdb_filepath) + f'_{chain_indic}.pdb' \
         if chain_indic not in pdb_filepath else db.get_pdb_name(pdb_filepath)
+    new_filepath = os.path.join(input_dataset_dir, "raw", pdb_code, filename)
     input_copy_cmd = f'cp {pdb_filepath} {os.path.join(input_dataset_dir, "raw", pdb_code, filename)}'
     input_copy_proc = subprocess.Popen(input_copy_cmd.split(), stdout=subprocess.PIPE, cwd=os.getcwd())
     _, _ = input_copy_proc.communicate()  # Wait until the input copy cmd is finished
+    return new_filepath
 
 
 def make_dataset(input_dataset_dir='datasets/Input/raw', output_dir='datasets/Input/interim', num_cpus=1,
@@ -618,6 +622,41 @@ def make_dataset(input_dataset_dir='datasets/Input/raw', output_dir='datasets/In
     pair.all_complex_to_pairs(complexes, source_type, get_pairs, pairs_dir, num_cpus)
 
 
+def recover_any_missing_chain_ids(interim_dataset_dir: str, new_pdb_filepath: str,
+                                  orig_pdb_filepath: str, pdb_code: str, chain_number: int):
+    """Restore any missing chain IDs for the chain represented by the corresponding Pandas DataFrame."""
+    orig_pdb_chain_id = '_'  # Default value for missing chain IDs
+    new_pdb_code = db.get_pdb_code(new_pdb_filepath)
+    orig_pdb_name = db.get_pdb_name(orig_pdb_filepath)
+    orig_pdb_df = PandasPdb().read_pdb(new_pdb_filepath).df['ATOM']
+    unique_chain_ids = np.unique(orig_pdb_df['chain_id'].values)
+    """Ascertain the chain ID corresponding to the original PDB file, using one of two available methods.
+      Method 1: Used with datasets such as EVCoupling adopting .atom filename extensions (e.g., 4DI3C.atom)
+      Method 2: Used with datasets such as DeepHomo adopting regular .pdb filename extensions (e.g., 2FNUA.pdb)"""
+    if len(unique_chain_ids) == 1 and unique_chain_ids[0].strip() == '':  # Method 1: Try to use filename differences
+        # No chain IDs were found, so we instead need to look to the original PDB filename to get the orig. chain ID
+        pdb_code_diffs = difflib.ndiff(new_pdb_code, orig_pdb_name)
+        for i, s in enumerate(pdb_code_diffs):
+            if s[0] == '+':
+                orig_pdb_chain_id = s[1:].strip()[0]
+                break
+    else:  # Method 2: Try to use unique chain IDs
+        # Assume the first/second index is the first non-empty chain ID (e.g., 'A')
+        orig_pdb_chain_id = unique_chain_ids[0] if (unique_chain_ids[0] != '') else unique_chain_ids[1]
+    # Update the existing Pair to contain the newly-recovered chain ID
+    pair_dir = os.path.join(interim_dataset_dir, 'pairs', pdb_code)
+    pair_filenames = [os.path.join(pair_dir, filename) for filename in os.listdir(pair_dir) if new_pdb_code in filename]
+    # Load in the existing Pair
+    with open(pair_filenames[0], 'rb') as f:
+        pair = dill.load(f)
+    # Update the corresponding chain ID
+    pair.df0.chain = orig_pdb_chain_id if chain_number == 1 else pair.df0.chain
+    pair.df1.chain = orig_pdb_chain_id if chain_number == 2 else pair.df1.chain
+    # Save the updated Pair
+    with open(pair_filenames[0], 'wb') as f:
+        dill.dump(pair, f)
+
+
 def generate_psaia_features(psaia_dir='~/Programs/PSAIA_1.0_source/bin/linux/psa',
                             psaia_config='datasets/builder/psaia_config_file_input.txt',
                             pdb_dataset='datasets/Input/raw', pkl_dataset='datasets/Input/interim/parsed',
@@ -727,9 +766,13 @@ def convert_input_pdb_files_to_pair(left_pdb_filepath: str, right_pdb_filepath:
     pdb_code = db.get_pdb_group(list(ca.get_complex_pdb_codes([left_pdb_filepath, right_pdb_filepath]))[0])
     # Iteratively execute the PDB file feature generation process
     create_input_dir_struct(input_dataset_dir, pdb_code)
-    copy_input_to_raw_dir(input_dataset_dir, left_pdb_filepath, pdb_code, 'l_u')
-    copy_input_to_raw_dir(input_dataset_dir, right_pdb_filepath, pdb_code, 'r_u')
+    new_l_u_filepath = copy_input_to_raw_dir(input_dataset_dir, left_pdb_filepath, pdb_code, 'l_u')
+    new_r_u_filepath = copy_input_to_raw_dir(input_dataset_dir, right_pdb_filepath, pdb_code, 'r_u')
     make_dataset(os.path.join(input_dataset_dir, 'raw'), os.path.join(input_dataset_dir, 'interim'))
+    recover_any_missing_chain_ids(os.path.join(input_dataset_dir, 'interim'),
+                                  new_l_u_filepath, left_pdb_filepath, pdb_code, 1)
+    recover_any_missing_chain_ids(os.path.join(input_dataset_dir, 'interim'),
+                                  new_r_u_filepath, right_pdb_filepath, pdb_code, 2)
     generate_psaia_features(psaia_dir=psaia_dir,
                             psaia_config=psaia_config,
                             pdb_dataset=os.path.join(input_dataset_dir, 'raw'),

requirements.txt

@@ -3,6 +3,7 @@ setuptools==57.4.0
 atom3-py3==0.1.9.8
 click==8.0.1
 easy-parallel-py3==0.1.6.4
+biopandas==0.2.9
 dill==0.3.4
 tqdm==4.62.0
 Sphinx==4.0.1

setup.py

@@ -4,7 +4,7 @@
 
 setup(
     name='DeepInteract',
-    version='1.0.2',
+    version='1.0.3',
     description='A geometric deep learning pipeline for predicting protein interface contacts.',
     author='Alex Morehead',
     author_email='[email protected]',
@@ -15,6 +15,7 @@
         'atom3-py3==0.1.9.8',
         'click==8.0.1',
         'easy-parallel-py3==0.1.6.4',
+        'biopandas==0.2.9',
         'dill==0.3.4',
         'tqdm==4.62.0',
         'Sphinx==4.0.1',