initial commit

.gitattributes

@@ -57,3 +57,12 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+davis.csv filter=lfs diff=lfs merge=lfs -text
+kiba.csv filter=lfs diff=lfs merge=lfs -text
+pdbbind-2020-refined.csv filter=lfs diff=lfs merge=lfs -text
+bindingdb-ic50.csv filter=lfs diff=lfs merge=lfs -text
+bindingdb-kd.csv filter=lfs diff=lfs merge=lfs -text
+davis-filter.csv filter=lfs diff=lfs merge=lfs -text
+bindingdb-ki.csv filter=lfs diff=lfs merge=lfs -text
+glaser.csv filter=lfs diff=lfs merge=lfs -text
+pdbbind-2020-combined.csv filter=lfs diff=lfs merge=lfs -text

bindingdb-ic50.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25a6488d678b3a1a2b83d78ab900fe5c49b96ad6b116efde158b5c5bdd39b557
+size 899074544

bindingdb-kd.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e18aaba78a1e2c0e9584fbc831954312d65e02deca70ce01b7514b5004a4c216
+size 57281302

bindingdb-ki.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8b08f64f5954c6189561014f802c103a86bb78d6644460e3f6f31697376a1f6
+size 259031100

data_sources.md

@@ -0,0 +1,9 @@
+davis-filter.csv: https://www.kaggle.com/datasets/christang0002/davis-and-kiba
+bindingdb-ic50.csv: https://tdcommons.ai/ (tdc python package)
+bindingdb-kd.csv: https://tdcommons.ai/ (tdc python package)
+bindingdb-ki.csv: https://tdcommons.ai/ (tdc python package)
+davis.csv: https://tdcommons.ai/ (tdc python package)
+kiba.csv: https://tdcommons.ai/ (tdc python package)
+pdbbind-2020-combined.csv: https://www.pdbbind.org.cn/
+pdbbind-2020-refined.csv: https://www.pdbbind.org.cn/
+glaser.csv: https://huggingface.co/datasets/jglaser/binding_affinity

davis-filter.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:484eaefff41f82de0f39811b589db7ed390061c952bc4199353d9e1c3700cfff
+size 8948685

davis.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34f698180a386c92ecc160da984d9d2c31bf1d153cb26a2d91d8eab4d7ec5fd7
+size 22441866

glaser.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73550c83f7fc9315bbcfa363304231f00f88dbd503222d241a47c8551ef65cd8
+size 1433278852

kiba.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56ac9893cbdc98ddc137a7ef4f5a5f00756da6ec5792b7ce9b2ccd270e4244d2
+size 102460723

pdbbind-2020-combined.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7221192655553cdfd07df4a71386aad31873bb8e0f27ec7fe26536a56a9df868
+size 11802666

pdbbind-2020-refined.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46ef8fb491e31ca6b9e6256742300b14d0cc545d0ebdf83cdd097155f98bbd99
+size 3025349

standardize_data.py

@@ -0,0 +1,205 @@
+import pandas as pd
+from rdkit import Chem
+from rdkit.Chem import MolToSmiles
+from scipy.stats import zscore
+from tqdm import tqdm
+import numpy as np
+
+def load_binding_affinity_dataset(csv_path, 
+                                  protein_col_idx, 
+                                  smiles_col_idx, 
+                                  affinity_col_idx, 
+                                  is_log10_affinity=True, 
+                                  canonicalize_smiles=True, 
+                                  affinity_unit="uM",
+                                  delimiter=','):
+    """
+    Load a protein-ligand binding affinity dataset and preprocess it.
+
+    Args:
+        csv_path (str): Path to the CSV file.
+        protein_col_idx (int): Column index containing protein sequences.
+        smiles_col_idx (int): Column index containing molecule SMILES.
+        affinity_col_idx (int): Column index containing binding affinities.
+        is_log10_affinity (bool): Whether affinities are in log10. Default is True.
+        canonicalize_smiles (bool): Whether to canonicalize SMILES. Default is True.
+        delimiter (str): Delimiter for the CSV file. Default is ','.
+
+    Returns:
+        pd.DataFrame: Processed DataFrame with columns "seq", "smiles_can",
+                      "affinity_uM", "neg_log10_affinity_M", and "affinity_norm".
+    """
+    # Load dataset
+    df = pd.read_csv(csv_path, delimiter=delimiter)
+
+    # Extract relevant columns
+    df = df.iloc[:, [protein_col_idx, smiles_col_idx, affinity_col_idx]]
+    df.columns = ["seq", "smiles", "affinity"]
+
+    # Canonicalize SMILES
+    if canonicalize_smiles:
+        def canonicalize(smiles):
+            try:
+                mol = Chem.MolFromSmiles(smiles)
+                return MolToSmiles(mol, canonical=True) if mol else None
+            except:
+                return None
+
+        from tqdm import tqdm
+        tqdm.pandas()
+        df["smiles_can"] = df["smiles"].progress_apply(canonicalize)
+        df = df[df["smiles_can"].notna()]
+    else:
+        df["smiles_can"] = df["smiles"]
+
+    # Process affinities
+    if not is_log10_affinity:
+        # Convert plain Kd value to neg log10(M)
+        df["affinity_uM"] = df["affinity"]/(1e3 if affinity_unit == "nM" else 1)
+        
+        df["neg_log10_affinity_M"] = -df["affinity_uM"].apply(lambda x: np.log10(x/1e6) if x > 0 else np.nan)
+    else:
+        # Convert log10 values to plain uM for clarity
+        df["neg_log10_affinity_M"] = df["affinity"]
+        df["affinity_uM"] = df["neg_log10_affinity_M"].apply(lambda x: (10**(-x))*1e6)
+
+    df.dropna(inplace=True)
+
+    # Z-score normalization
+    df["affinity_norm"] = zscore(df["neg_log10_affinity_M"])
+
+    # Select and reorder columns
+    df = df[["seq", "smiles_can", "affinity_uM", "neg_log10_affinity_M", "affinity_norm"]]
+
+    # Add normalization parameters as columns for reference
+    df["affinity_mean"] = df["neg_log10_affinity_M"].mean()
+    df["affinity_std"] = df["neg_log10_affinity_M"].std()
+
+    return df.sort_values(by="affinity_norm", ascending=False)
+
+dataset = load_binding_affinity_dataset(
+    csv_path="data/raw_data/bindingdb_ic50.csv",
+    protein_col_idx=3,
+    smiles_col_idx=1,
+    affinity_col_idx=4,
+    is_log10_affinity=False,  # Specify if Kd values are plain
+    canonicalize_smiles=True,
+    affinity_unit="nM",
+    delimiter=","
+)
+dataset.to_csv("data/bindingdb-ic50.csv", index=False)
+
+# View processed dataset
+print(dataset.head())
+
+
+# Code for loading tdc data:
+# import pandas as pd
+# from rdkit import Chem
+# from tdc.multi_pred import DTI
+
+# def process_dataset(name):
+#     data = DTI(name=name)
+#     data.harmonize_affinities(mode='mean')
+#     data.convert_to_log()
+#     df = data.get_data()
+#     df['smiles_can'] = df['Drug'].apply(lambda s: Chem.MolToSmiles(Chem.MolFromSmiles(s), isomericSmiles=True, canonical=True) if Chem.MolFromSmiles(s) else None)
+#     return df[['smiles_can', 'Target', 'Y']].dropna(subset=['smiles_can']).rename(columns={'Target': 'seq', 'Y': 'neg_log_10_affinity'})
+
+# datasets = ['BindingDB_Ki', 'BindingDB_Kd', 'BindingDB_IC50', 'DAVIS', 'KIBA']
+# processed_data = [process_dataset(name) for name in datasets]
+
+# binding_db = pd.concat(processed_data[:3]).drop_duplicates().reset_index(drop=True)
+# binding_db.to_csv("data/bindingdb.csv", index=False)
+# processed_data[3].to_csv("data/davis.csv", index=False)
+# processed_data[4].to_csv("data/kiba.csv", index=False)
+
+# code for loading pdbbind data:
+# import os
+# from pathlib import Path
+# from Bio import PDB
+# from Bio.PDB.Polypeptide import PPBuilder
+# from rdkit import Chem
+# from rdkit.Chem import AllChem
+# import pandas as pd
+# from tqdm import tqdm
+
+# ppb = PPBuilder()
+
+# def get_protein_sequence(structure):
+#     """Extract protein sequence from a PDB structure."""
+#     sequence = ""
+#     for pp in ppb.build_peptides(structure):
+#         sequence += str(pp.get_sequence())
+#     return sequence
+
+# def get_canonical_smiles(mol):
+#     """Convert RDKit molecule to canonical SMILES."""
+#     return Chem.MolToSmiles(mol, isomericSmiles=True, canonical=True)
+
+# def process_pdbbind_data(pdbbind_dir, index_file):
+#     pdbbind_dir = Path(pdbbind_dir).expanduser()
+#     parser = PDB.PDBParser(QUIET=True)
+#     data = []
+
+#     # Read the index file
+#     df_index = pd.read_csv(index_file, sep='\s+', header=None, comment= "#", usecols=[0,1,2,3,4,6,7],
+#                            names=['PDB_ID', 'Resolution', 'Release_Year', '-logKd/Ki', 'Kd/Ki', 'Reference', 'Ligand_Name'])
+
+#     # Get the total number of entries for progress tracking
+#     total_entries = len(df_index)
+
+#     # Use tqdm for progress tracking
+#     with tqdm(total=total_entries, desc="Processing PDBbind data") as pbar:
+#         for _, row in df_index.iterrows():
+#             pdb_id = row['PDB_ID']
+#             subdir = pdbbind_dir / pdb_id
+
+#             if subdir.is_dir():
+#                 # Process protein
+#                 protein_file = subdir / f"{pdb_id}_protein.pdb"
+#                 if protein_file.exists():
+#                     structure = parser.get_structure(pdb_id, protein_file)
+#                     sequence = get_protein_sequence(structure)
+
+#                 # Process ligand
+#                 ligand_file = subdir / f"{pdb_id}_ligand.mol2"
+#                 if ligand_file.exists():
+#                     mol = Chem.MolFromMol2File(str(ligand_file))
+#                     if mol is not None:
+#                         smiles = get_canonical_smiles(mol)
+
+#                         # Get binding affinity
+#                         neg_log_10_affinity = row['-logKd/Ki']
+
+#                         # Add to data list
+#                         data.append({
+#                             'smiles_can': smiles,
+#                             'seq': sequence,
+#                             'neg_log_10_affinity_M': neg_log_10_affinity
+#                         })
+
+#             pbar.update(1)
+
+#     return pd.DataFrame(data)
+
+# # Process data from PDBbind refined set
+# pdbbind_refined_dir = "~/Data/PDBBind/PDBbind_v2020_refined"
+# index_refined_file = "/home/tyler/Data/PDBBind/index/INDEX_refined_data.2020"
+# df_refined = process_pdbbind_data(pdbbind_refined_dir, index_refined_file)
+
+# # Process data from PDBbind general set
+# pdbbind_general_dir = "~/Data/PDBBind/PDBbind_v2020_other_PL"
+# index_general_file = "/home/tyler/Data/PDBBind/index/INDEX_general_PL_data.2020"
+# df_general = process_pdbbind_data(pdbbind_general_dir, index_general_file)
+
+# # Combine dataframes
+# df_combined = pd.concat([df_refined, df_general], ignore_index=True)
+
+# # Remove duplicates (if any) and reset index
+# df_combined = df_combined.drop_duplicates().reset_index(drop=True)
+
+# # Save to CSV
+# output_file = "data/pdbbind_2020_combined.csv"
+# df_combined.to_csv(output_file, index=False)
+# print(f"Saved {len(df_combined)} entries to {output_file}")