Datasets:
File size: 14,286 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 | import torch
import torch.nn as nn
import torch.nn.functional as F
from model.molas_blocks import *
from torch.nn import Linear, ReLU, Dropout, Sequential, BatchNorm1d
class MolAS(nn.Module):
def __init__(self, num_node_features=960, num_ligand_features=384, num_classes=8,
num_res_blocks=4, res_dim=128, hidden_res_dim=256, dropout_rate=0.3,
norm_type='graph', aggr_type='attentional', num_heads_protein=1, num_heads_ligand=2):
super(MolAS, self).__init__()
self.num_res_blocks = num_res_blocks
self.num_ligand_features = num_ligand_features
self.pool1 = MultiAttentionAggr(num_node_features, num_heads=num_heads_protein)
# Residual decoder
self.Res_decoder = nn.ModuleDict()
self.Res_decoder['res_projection'] = nn.Linear(num_node_features * num_heads_protein + num_ligand_features, res_dim)
for i in range(num_res_blocks):
self.Res_decoder[f'res_block{i+1}'] = ResidualBlock(res_dim, hidden_res_dim, dropout_rate=dropout_rate)
self.regressor = nn.Linear(res_dim * (1 + num_res_blocks), num_classes)
def forward(self, ligand_data, protein_data, return_features=False):
# Graph embeddings
x1 = self.pool1(protein_data.x, protein_data.batch)
# x2 = self.pool2(ligand_data.x, ligand_data.batch)
x2 = ligand_data.x.view(-1, self.num_ligand_features)
if return_features:
return torch.cat((x1, x2), dim=1)
x = self.Res_decoder['res_projection'](torch.cat((x1, x2), dim=1))
# Apply residual blocks (no checkpointing to avoid determinism issues)
res_outputs = [x]
for i in range(self.num_res_blocks):
res_out = self.Res_decoder[f'res_block{i+1}'](res_outputs[-1])
res_outputs.append(res_out)
x = torch.cat(res_outputs, dim=1)
return self.regressor(x)
class MolAS_GCN_GAT_GINE(nn.Module):
def __init__(self, num_node_features=960, num_ligand_features=384, num_classes=8,
num_res_blocks=4, res_dim=128, hidden_res_dim=256, dropout_rate=0.3,
norm_type='graph', aggr_type='attentional', num_heads_protein=1, num_heads_ligand=2):
super(MolAS_GCN_GAT_GINE, self).__init__()
self.num_res_blocks = num_res_blocks
self.num_ligand_features = num_ligand_features
# GNN encoders
self.GCN_encoder = GCN(num_node_features=num_node_features, num_layers=3, hidden_dim=128,
norm_type=norm_type, aggr_type=aggr_type)
self.GAT_encoder = GAT(num_node_features=num_node_features, num_layers=3, hidden_dims=[128, 64, 32], heads=4,
norm_type=norm_type, aggr_type=aggr_type)
self.GIN_encoder = GIN(num_node_features=num_node_features, num_layers=3, hidden_dims=[128, 64, 32],
norm_type=norm_type, aggr_type=aggr_type)
out_dim = self.GCN_encoder.out_dim + self.GAT_encoder.out_dim + self.GIN_encoder.out_dim + self.num_ligand_features
# Residual decoder
self.Res_decoder = nn.ModuleDict()
self.Res_decoder['res_projection'] = nn.Linear(out_dim, res_dim)
for i in range(num_res_blocks):
self.Res_decoder[f'res_block{i+1}'] = ResidualBlock(res_dim, hidden_res_dim, dropout_rate=dropout_rate)
self.regressor = nn.Linear(res_dim * (1 + num_res_blocks), num_classes)
def forward(self, ligand_data, protein_data, return_features=False):
# Graph embeddings
x_gcn = F.normalize(self.GCN_encoder(protein_data), dim=-1)
x_gat = F.normalize(self.GAT_encoder(protein_data), dim=-1)
x_gin = F.normalize(self.GIN_encoder(protein_data), dim=-1)
# x2 = self.pool2(ligand_data.x, ligand_data.batch)
x_ligand = ligand_data.x.view(-1, self.num_ligand_features)
if return_features:
return torch.cat((x_gcn, x_gat, x_gin, x_ligand), dim=1)
x = self.Res_decoder['res_projection'](torch.cat((x_gcn, x_gat, x_gin, x_ligand), dim=1))
# Apply residual blocks (no checkpointing to avoid determinism issues)
res_outputs = [x]
for i in range(self.num_res_blocks):
res_out = self.Res_decoder[f'res_block{i+1}'](res_outputs[-1])
res_outputs.append(res_out)
x = torch.cat(res_outputs, dim=1)
return self.regressor(x)
class MolAS_EGNN_GAT_GINE(nn.Module):
def __init__(self, num_node_features=960, num_ligand_features=384, num_classes=8,
num_res_blocks=4, res_dim=128, hidden_res_dim=256, dropout_rate=0.3,
norm_type='graph', aggr_type='attentional', num_heads_protein=1, num_heads_ligand=2):
super(MolAS_EGNN_GAT_GINE, self).__init__()
self.num_res_blocks = num_res_blocks
self.num_ligand_features = num_ligand_features
# GNN encoders
self.EGNN_encoder = EGNN(num_node_features=num_node_features, num_layers=3, hidden_dim=128,
norm_type=norm_type, aggr_type=aggr_type)
self.GAT_encoder = GAT(num_node_features=num_node_features, num_layers=3, hidden_dims=[128, 64, 32], heads=4,
norm_type=norm_type, aggr_type=aggr_type)
self.GIN_encoder = GIN(num_node_features=num_node_features, num_layers=3, hidden_dims=[128, 64, 32],
norm_type=norm_type, aggr_type=aggr_type)
out_dim = self.EGNN_encoder.out_dim + self.GAT_encoder.out_dim + self.GIN_encoder.out_dim + self.num_ligand_features
# Residual decoder
self.Res_decoder = nn.ModuleDict()
self.Res_decoder['res_projection'] = nn.Linear(out_dim, res_dim)
for i in range(num_res_blocks):
self.Res_decoder[f'res_block{i+1}'] = ResidualBlock(res_dim, hidden_res_dim, dropout_rate=dropout_rate)
self.regressor = nn.Linear(res_dim * (1 + num_res_blocks), num_classes)
def forward(self, ligand_data, protein_data, return_features=False):
# Graph embeddings
x_egnn = F.normalize(self.EGNN_encoder(protein_data), dim=-1)
x_gat = F.normalize(self.GAT_encoder(protein_data), dim=-1)
x_gin = F.normalize(self.GIN_encoder(protein_data), dim=-1)
# x2 = self.pool2(ligand_data.x, ligand_data.batch)
x_ligand = ligand_data.x.view(-1, self.num_ligand_features)
if return_features:
return torch.cat((x_egnn, x_gat, x_gin, x_ligand), dim=1)
x = self.Res_decoder['res_projection'](torch.cat((x_egnn, x_gat, x_gin, x_ligand), dim=1))
# Apply residual blocks (no checkpointing to avoid determinism issues)
res_outputs = [x]
for i in range(self.num_res_blocks):
res_out = self.Res_decoder[f'res_block{i+1}'](res_outputs[-1])
res_outputs.append(res_out)
x = torch.cat(res_outputs, dim=1)
return self.regressor(x)
class MolASGT(nn.Module):
def __init__(self, num_node_features=960, num_ligand_features=384, num_classes=8,
num_res_blocks=4, res_dim=128, hidden_res_dim=256, dropout_rate=0.3,
norm_type='graph', aggr_type='attentional', num_heads_protein=1, num_heads_ligand=2):
super(MolASGT, self).__init__()
self.num_res_blocks = num_res_blocks
self.num_ligand_features = num_ligand_features
self.gt = GT(num_node_features=num_node_features, num_layers=3, hidden_dims=[128, 64, 32],
norm_type=norm_type, aggr_type=aggr_type)
out_dim = self.gt.out_dim + self.num_ligand_features
# Residual decoder
self.Res_decoder = nn.ModuleDict()
self.Res_decoder['res_projection'] = nn.Linear(out_dim, res_dim)
for i in range(num_res_blocks):
self.Res_decoder[f'res_block{i+1}'] = ResidualBlock(res_dim, hidden_res_dim, dropout_rate=dropout_rate)
self.regressor = nn.Linear(res_dim * (1 + num_res_blocks), num_classes)
def forward(self, ligand_data, protein_data, return_features=False):
# Graph embeddings
x_gt = self.gt(protein_data)
# x2 = self.pool2(ligand_data.x, ligand_data.batch)
x2 = ligand_data.x.view(-1, self.num_ligand_features)
if return_features:
return torch.cat((x_gt, x2), dim=1)
x = self.Res_decoder['res_projection'](torch.cat((x_gt, x2), dim=1))
# Apply residual blocks (no checkpointing to avoid determinism issues)
res_outputs = [x]
for i in range(self.num_res_blocks):
res_out = self.Res_decoder[f'res_block{i+1}'](res_outputs[-1])
res_outputs.append(res_out)
x = torch.cat(res_outputs, dim=1)
return self.regressor(x)
class MolASMinimal(nn.Module):
def __init__(self, num_node_features=1152, num_classes=8, num_ligand_features=384,
num_res_blocks=3, res_dim=128, hidden_res_dim=256, dropout_rate=0.3, num_heads_protein=1,
aggr_type='attentional'):
super(MolASMinimal, self).__init__()
self.num_res_blocks = num_res_blocks
self.num_res_blocks = num_res_blocks
self.num_ligand_features = num_ligand_features
self.pool1 = MultiAttentionAggr(num_node_features, num_heads=num_heads_protein)
self.processor = Sequential(
Linear(num_node_features + num_ligand_features, res_dim),
ReLU(),
Dropout(dropout_rate),
nn.Linear(res_dim, num_classes)
)
def forward(self, ligand_data, protein_data, return_features=False):
x1 = self.pool1(protein_data.x, protein_data.batch)
x2 = ligand_data.x.view(-1, self.num_ligand_features)
if return_features:
return torch.cat((x1, x2), dim=1)
x = self.processor(torch.cat((x1, x2), dim=1))
return x
class MCGNNASDock(nn.Module):
def __init__(self, num_node_features=1152, num_node_features_ligand=25, num_classes=8, dropout_rate=0.3, input_dropout_rate=0.1):
super(MCGNNASDock, self).__init__()
self.model1 = GAT_L_NO_FIXED_OUTPUT_DIM(num_node_features=num_node_features_ligand, num_classes=num_classes)
self.model2 = GCN_GAT_GINE_FIXED_OUT_DIM(num_node_features=num_node_features, num_classes=num_classes)
""" No Residual connections
self.combiner = nn.Sequential(
# We move the concatenation to the combiner
# 3 linear layers with dropout and ReLU activation
nn.Linear(320 + 96, 256), # 320 + 96 -> 128 + 128
nn.ReLU(), ## nn.ReLU,
nn.Dropout(dropout_rate),
nn.Linear(256, 128),
nn.ReLU(),
nn.Dropout(dropout_rate),
nn.Linear(128, 64),
nn.ReLU(),
nn.Dropout(dropout_rate),
nn.Linear(64, 32),
nn.ReLU(),
nn.Dropout(dropout_rate),
nn.Linear(32, 16),
nn.ReLU(), ## nn.Sigmoid
nn.Dropout(dropout_rate),
nn.Linear(16, num_classes),
nn.Sigmoid()
)
"""
# MLP with residual connections
# self.input_projection = nn.Linear(320 + 96, 256) # Project to initial hidden size
# Residual blocks
# self.res_block1 = ResidualBlock(256, 512, dropout_rate)
# self.fc1 = nn.Linear(256, 128)
self.res_block20 = ResidualBlock(128, 256, dropout_rate=0.3)
self.res_block21 = ResidualBlock(128, 256, dropout_rate=0.3)
self.res_block22 = ResidualBlock(128, 256, dropout_rate=0.3)
# Calculate dense connection input sizes
# After 3 res blocks: 128 + 128 + 128 + 128 = 512
self.fc2 = nn.Linear(512, 64) # Updated from 128 to 512
self.res_block30 = ResidualBlock(64, 128, dropout_rate=0.3)
self.res_block31 = ResidualBlock(64, 128, dropout_rate=0.3)
self.res_block32 = ResidualBlock(64, 128, dropout_rate=0.3)
# After next 3 res blocks: 64 + 64 + 64 + 64 = 256
self.fc3 = nn.Linear(256, 32) # Updated from 64 to 256
# Add batch normalization
# self.bn1 = nn.BatchNorm1d(256)
# self.bn2 = nn.BatchNorm1d(128)
self.bn3 = nn.BatchNorm1d(64)
self.bn4 = nn.BatchNorm1d(32)
self.relu = nn.ReLU()
self.dropout = nn.Dropout(dropout_rate)
# self.input_dropout = nn.Dropout(input_dropout_rate)
# Final classifier
self.classifier = nn.Linear(32, num_classes)
"""No Residual connections
def forward(self, data1, data2):
# Get predictions from both models
output1 = self.model1(data1)
output2 = self.model2(data2)
# Concatenate the outputs from both models
combined_output = torch.cat((output1, output2), dim=1)
# Apply the combiner to the concatenated outputs
final_output = self.combiner(combined_output)
return final_output
"""
def forward(self, data1, data2):
# Get predictions from both models
output1 = self.model1(data1)
output2 = self.model2(data2)
# Concatenate the outputs from both models
combined_output = torch.cat((output1, output2), dim=1)
# Apply residual blocks with dense connections
x1 = self.res_block20(combined_output)
x2 = self.res_block21(x1)
x3 = self.res_block22(x2)
# Dense connection: concatenate all previous block outputs
dense_input = torch.cat((combined_output, x1, x2, x3), dim=1)
x = self.fc2(dense_input)
x = self.bn3(x)
x = self.relu(x)
x = self.dropout(x)
x4 = self.res_block30(x)
x5 = self.res_block31(x4)
x6 = self.res_block32(x5)
# Another dense connection
dense_input2 = torch.cat((x, x4, x5, x6), dim=1)
x = self.fc3(dense_input2)
x = self.bn4(x)
x = self.relu(x)
x = self.dropout(x)
# Final classification
final_output = self.classifier(x)
return final_output |