Datasets:
File size: 14,859 Bytes
84a242a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 | import torch
from torch_geometric.data import Data, Dataset
from torch_geometric.loader import DataLoader
import pandas as pd
import math
import numpy as np
from pathlib import Path
#---------------------------Change dataset-----------------------------
# These will be set dynamically via function parameters
path = '/data1/home/jw1017/AS_Dock/data/MolAS'
ligand_graph_path = f'{path}/ligand_graphs'
protein_graph_path = f'{path}/protein_graphs_esmc_600m'
#----------------------------------------------------------------------
def resolve_matrix_path(benchmark='posex_self_docking', matrix_name='pb_ratio'):
data_root = Path(path)
candidates = [
data_root / f'{benchmark}_{matrix_name}_trimmed.csv',
data_root / f'{benchmark}_{matrix_name}_sorted.csv',
data_root / f'{benchmark}_{matrix_name}.csv',
]
for candidate in candidates:
if candidate.exists():
return str(candidate)
return str(candidates[0])
def get_dataset_paths(benchmark='posex_self_docking'):
"""Generate dataset paths based on benchmark and relaxation settings"""
pb_ratios_path = resolve_matrix_path(benchmark, 'pb_ratio')
rmsd_path = resolve_matrix_path(benchmark, 'rmsd')
return pb_ratios_path, rmsd_path
def get_score_path(benchmark='posex_self_docking'):
return resolve_matrix_path(benchmark, 'score')
def normalize_metric_column(column_name):
suffixes = [
'_strict_score_from_rebuilt_raw',
'_pred_strict_score',
'_strict_score',
'_pred_pb_ratio',
'_pred_rmsd',
'_pb_ratio',
'_rmsd',
'_score',
]
for suffix in suffixes:
if column_name.endswith(suffix):
return column_name[:-len(suffix)]
return column_name
def split_pdb_ccd_id_column(pdb_ccd_ids):
protein = [value.split('_')[0] for value in pdb_ccd_ids]
ligand = ['_'.join(value.split('_')[1:]) for value in pdb_ccd_ids]
return protein, ligand
def build_label_dataframe(matrix_df, incl_columns:list = []):
renamed_columns = {
col: normalize_metric_column(col)
for col in matrix_df.columns
if col != 'PDB_CCD_ID'
}
matrix_df = matrix_df.rename(columns=renamed_columns)
protein, ligand = split_pdb_ccd_id_column(matrix_df['PDB_CCD_ID'])
final_df = matrix_df.copy()
final_df.insert(0, 'protein', protein)
final_df.insert(1, 'ligand', ligand)
final_df = final_df.drop(columns=['PDB_CCD_ID'])
if len(incl_columns) > 0:
requested = incl_columns + ['protein', 'ligand']
missing = [col for col in incl_columns if col not in final_df.columns]
if missing:
raise ValueError(
f"Requested incl_columns missing from matrix: {missing}. "
f"Available columns: {list(final_df.columns[2:])}"
)
final_df = final_df[final_df.columns.intersection(requested)]
label = [final_df.iloc[i, 2:].values for i in range(len(final_df))]
final_df['label'] = label
num_classes = len(final_df.columns[2:-1])
assert not final_df.isnull().values.any(), "DataFrame contains NaN values."
assert not final_df.isin([float('inf'), float('-inf')]).values.any(), "DataFrame contains infinite values."
assert all(isinstance(t, (list, np.ndarray)) for t in final_df['label']), "label column contains non-list/array values."
assert all(len(t) == num_classes for t in final_df['label']), "Inconsistent label lengths."
return final_df, num_classes
def pb_score(x):
# binary score
return 1 if x == 1 else 0
# RMSD scores
def rmsd_score(x, lambda_rmsd=3):
M = 2
if x > 5 or x < 0:
return 0
return (1 + math.exp(-lambda_rmsd * M)) / (1 + math.exp(lambda_rmsd * (x - M)))
# Dataset classes
# ligand graph
class SMILESDataset(Dataset):
def __init__(self, dataframe):
self.dataframe = dataframe
self.ligand_graph_path = ligand_graph_path
def __len__(self):
return len(self.dataframe)
def __getitem__(self, idx):
ligand_name = self.dataframe.iloc[idx]['ligand']
protein_name = self.dataframe.iloc[idx]['protein']
names = f'{protein_name}_{ligand_name}'
# Direct loading without cache for better performance
graph_path = f'{self.ligand_graph_path}/pyg_graph_{names}.pt'
nodes, edge_index, edge_attr, pos = torch.load(graph_path, weights_only=False, map_location='cpu')
# Create data with explicit memory optimization
graph_data = Data(
x=nodes[1].to(torch.float32).contiguous(), # Ensure contiguous memory layout and dtype
edge_index=edge_index[1].contiguous(),
edge_attr=edge_attr[1].to(torch.float32).contiguous(),
pos=pos[1].to(torch.float32).contiguous()
)
graph_data.names = names
return graph_data
# Dataset classes
# ligand graph
class SMILESDatasetChemberta(Dataset):
def __init__(self, dataframe):
self.dataframe = dataframe
lig_chemberta = torch.load(f"{path}/ligand_chemberta.pt")
self.lig_emb_map = dict(zip(lig_chemberta["ligand_ids"],
lig_chemberta["embeddings"]))
def __len__(self):
return len(self.dataframe)
def __getitem__(self, idx):
ligand_name = self.dataframe.iloc[idx]['ligand']
protein_name = self.dataframe.iloc[idx]['protein']
names = f'{protein_name}_{ligand_name}'
# Create data with explicit memory optimization
graph_data = Data(
x=self.lig_emb_map[names].to(torch.float32).contiguous()
)
graph_data.names = names
return graph_data
# protein graph
class GraphDataset(Dataset):
def __init__(self, dataframe):
self.dataframe = dataframe
self.protein_graph_path = protein_graph_path
# Cache for loaded graphs to prevent memory leaks
self._graph_cache = {}
def __len__(self):
return len(self.dataframe)
def validate_graph(self, x, pos, edge_index, edge_attr=None):
N = pos.size(0)
assert x.size(0) == N
if edge_index.numel():
m, n = int(edge_index.max()), int(edge_index.min())
assert 0 <= n and m < N
if edge_attr is not None:
assert edge_attr.size(0) == edge_index.size(1)
def __getitem__(self, idx):
protein_name = self.dataframe.iloc[idx]['protein']
ligand_name = self.dataframe.iloc[idx]['ligand']
names = f'{protein_name}_{ligand_name}'
# Use cache to avoid reloading the same graph
if names not in self._graph_cache:
graph_path = f'{self.protein_graph_path}/pyg_graph_{names}_esmc_600m.pt'
# graph_path = f'{self.protein_graph_path}/pytorch_graph_{names}_protein.pt'
graph = torch.load(graph_path, weights_only=False)
x, pos, edge_index, edge_attr = graph.x, graph.pos, graph.edge_index, graph.edge_attr
x = x.float()
self._graph_cache[names] = (x, pos, edge_index, edge_attr)
else:
x, pos, edge_index, edge_attr = self._graph_cache[names]
x = x.float()
self.validate_graph(x, pos, edge_index, edge_attr)
# print(f'Shapes - x: {x.shape}, pos: {pos.shape}, edge_index: {edge_index.shape}, edge_attr: {edge_attr.shape}')
# print(f'Type - x: {x.dtype}, pos: {pos.dtype}, edge_index: {edge_index.dtype}, edge_attr: {edge_attr.dtype}')
# Create data with explicit memory optimization
graph_data = Data(
x=x.to(torch.float32).contiguous(), # Ensure contiguous memory layout and dtype
pos=pos.to(torch.float32).contiguous(),
edge_index=edge_index.contiguous(),
edge_attr=edge_attr.to(torch.float32).contiguous() # Slice and make contiguous
)
graph_data.names = names
return graph_data
class LabelDataset(Dataset):
def __init__(self, dataframe):
self.dataframe = dataframe.reset_index(drop=True)
def __len__(self):
return len(self.dataframe)
def __getitem__(self, idx):
label = self.dataframe.loc[idx, "label"].astype(float)
return torch.tensor(label, dtype=torch.float32)
# Create Combined Dataset for use in stacked model
class CombinedDataset(Dataset):
def __init__(self, dataframe):
self.smiles_dataset = SMILESDatasetChemberta(dataframe)
# self.smiles_dataset = SMILESDataset(dataframe)
self.graph_dataset = GraphDataset(dataframe)
self.label_dataset = LabelDataset(dataframe)
def __len__(self):
return len(self.smiles_dataset)
def __getitem__(self, idx):
# Get data from both datasets
smiles_data = self.smiles_dataset[idx]
graph_data = self.graph_dataset[idx]
label_data = self.label_dataset[idx]
# Since both dataframes are the same, ligand names should match
assert smiles_data.names == graph_data.names, "names do not match."
return smiles_data, graph_data, label_data
def prepare_data_composite(incl_columns:list = [], benchmark='posex_self_docking'):
# Load the 'pb_ratios' and 'rmsd' datasets
pb_ratios_path, rmsd_path = get_dataset_paths(benchmark)
pb_ratios = pd.read_csv(pb_ratios_path)
rmsd = pd.read_csv(rmsd_path)
# Merge into a single DataFrame
protein = [i.split('_')[0] for i in rmsd['PDB_CCD_ID']]
ligand = ['_'.join(i.split('_')[1:]) for i in rmsd['PDB_CCD_ID']]
rmsd.insert(0, 'protein', protein)
rmsd.insert(1, 'ligand', ligand)
rmsd = rmsd.drop(columns=['PDB_CCD_ID'])
rmsd.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-1]) for col_name in rmsd.columns[2:]]
pb_ratios.insert(0, 'protein', protein)
pb_ratios.insert(1, 'ligand', ligand)
pb_ratios = pb_ratios.drop(columns=['PDB_CCD_ID'])
pb_ratios.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-2]) for col_name in pb_ratios.columns[2:]]
pb_ratios = pb_ratios[rmsd.columns] # important: align columns
# Score computation
pb_scores = pb_ratios.iloc[:, 2:].map(pb_score)
rmsd_scores = rmsd.iloc[:, 2:].map(rmsd_score)
final_df = pb_scores * rmsd_scores
final_df.insert(0, 'protein', protein)
final_df.insert(1, 'ligand', ligand)
# Filter to keep only the specified columns(algorithms)
if len(incl_columns) > 0:
final_df = final_df[final_df.columns.intersection(incl_columns+['protein', 'ligand'])]
# final_df.iloc[:, 2:] = np.random.rand(final_df.shape[0], final_df.shape[1]-2) # Dummy features for testing
label = [final_df.iloc[i,2:].values for i in range(len(final_df))]
final_df['label'] = label
num_classes = len(final_df.columns[2:-1])
# Health check of the final dataframe
assert not final_df.isnull().values.any(), "DataFrame contains NaN values."
assert not final_df.isin([float('inf'), float('-inf')]).values.any(), "DataFrame contains infinite values."
assert all(isinstance(t, (list, np.ndarray)) for t in final_df['label']), "label column contains non-list/array values."
assert all(len(t) == num_classes for t in final_df['label']), "Inconsistent label lengths."
return final_df, num_classes
def prepare_data_score(incl_columns:list = [], benchmark='posex_self_docking', score_path=None):
if score_path is None:
score_path = get_score_path(benchmark)
score_df = pd.read_csv(score_path)
return build_label_dataframe(score_df, incl_columns)
# Directly return the RMSD scores; Currently used for evaluation only!
def prepare_data_rmsd(incl_columns:list = [], benchmark='posex_self_docking'):
# Load the 'pb_ratios' and 'rmsd' datasets
pb_ratios_path, rmsd_path = get_dataset_paths(benchmark)
pb_ratios = pd.read_csv(pb_ratios_path)
rmsd = pd.read_csv(rmsd_path)
# Merge into a single DataFrame
protein = [i.split('_')[0] for i in rmsd['PDB_CCD_ID']]
ligand = ['_'.join(i.split('_')[1:]) for i in rmsd['PDB_CCD_ID']]
rmsd.insert(0, 'protein', protein)
rmsd.insert(1, 'ligand', ligand)
rmsd = rmsd.drop(columns=['PDB_CCD_ID'])
rmsd.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-1]) for col_name in rmsd.columns[2:]]
pb_ratios.insert(0, 'protein', protein)
pb_ratios.insert(1, 'ligand', ligand)
pb_ratios = pb_ratios.drop(columns=['PDB_CCD_ID'])
pb_ratios.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-2]) for col_name in pb_ratios.columns[2:]]
pb_ratios = pb_ratios[rmsd.columns] # important: align columns
final_df = rmsd
# Filter to keep only the specified columns(algorithms)
if len(incl_columns) > 0:
final_df = final_df[final_df.columns.intersection(incl_columns+['protein', 'ligand'])]
label = [final_df.iloc[i,2:].values for i in range(len(final_df))]
final_df['label'] = label
num_classes = len(final_df.columns[2:-1])
return final_df, num_classes
# Directly return the PoseBuster pass (or not); Currently used for evaluation only!
def prepare_data_pose(incl_columns:list = [], benchmark='posex_self_docking'):
# Load the 'pb_ratios' and 'rmsd' datasets
pb_ratios_path, rmsd_path = get_dataset_paths(benchmark)
pb_ratios = pd.read_csv(pb_ratios_path)
rmsd = pd.read_csv(rmsd_path)
# Merge into a single DataFrame
protein = [i.split('_')[0] for i in rmsd['PDB_CCD_ID']]
ligand = ['_'.join(i.split('_')[1:]) for i in rmsd['PDB_CCD_ID']]
rmsd.insert(0, 'protein', protein)
rmsd.insert(1, 'ligand', ligand)
rmsd = rmsd.drop(columns=['PDB_CCD_ID'])
rmsd.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-1]) for col_name in rmsd.columns[2:]]
pb_ratios.insert(0, 'protein', protein)
pb_ratios.insert(1, 'ligand', ligand)
pb_ratios = pb_ratios.drop(columns=['PDB_CCD_ID'])
pb_ratios.columns=['protein', 'ligand'] + ['_'.join(col_name.split('_')[:-2]) for col_name in pb_ratios.columns[2:]]
pb_ratios = pb_ratios[rmsd.columns] # important: align columns
final_df = pb_ratios.iloc[:, 2:].map(pb_score).astype(int) # True if PoseBuster passes, False otherwise
final_df.insert(0, 'protein', protein)
final_df.insert(1, 'ligand', ligand)
# Filter to keep only the specified columns(algorithms)
if len(incl_columns) > 0:
final_df = final_df[final_df.columns.intersection(incl_columns+['protein', 'ligand'])]
label = [final_df.iloc[i,2:].values for i in range(len(final_df))]
final_df['label'] = label
num_classes = len(final_df.columns[2:-1])
return final_df, num_classes |