JESTR1/models.py at main · HassounLab/JESTR1 · GitHub

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
# -*- coding: utf-8 -*-
"""
Created on Thu Aug 17 10:18:01 2023

@author: apurv
"""
import torch
import torch.nn as nn
import dgl
from dgllife.model import GCN, GAT

class MolEnc(nn.Module):
    def __init__(self, params, in_dim):
        super(MolEnc, self).__init__()
        dropout = [params['gnn_dropout'] for _ in range(len(params['gnn_channels']))]
        batchnorm = [True for _ in range(len(params['gnn_channels']))]
        gnn_map = {
            "gcn": GCN(in_dim, params['gnn_channels'], batchnorm = batchnorm, dropout = dropout),
            "gat": GAT(in_dim, params['gnn_channels'], params['attn_heads'])
        }
        self.GNN = gnn_map[params['gnn_type']]
        self.pool = dgl.nn.pytorch.glob.MaxPooling()
        self.fc1_graph = nn.Linear(params['gnn_channels'][len(params['gnn_channels']) - 1], params['gnn_hidden_dim'] * 2)
        self.fc2_graph = nn.Linear(params['gnn_hidden_dim'] * 2, params['final_embedding_dim'])
        self.W_out1 = nn.Linear(params['final_embedding_dim'], params['gnn_hidden_dim'] * 2)
        self.W_out2 = nn.Linear(params['gnn_hidden_dim'] * 2, params['gnn_hidden_dim'])
        self.dropout = nn.Dropout(params['fc_dropout'])
        self.relu = nn.ReLU()

    def forward(self, g1, f1):
        f = self.GNN(g1, f1)
        h = self.pool(g1, f)
        h1 = self.relu(self.fc1_graph(h))
        h1 = self.dropout(h1)
        h1 = self.fc2_graph(h1)
        h1 = self.dropout(h1)
        #h1 = self.relu(h1)

        return h1

    def load_weights(self, pretrained_model):
        # load pretrained_model weights
        state_dict = {}
        loaded_dict = torch.load(pretrained_model, map_location=torch.device('cpu'))
        if loaded_dict.get('model_state_dict', None) is not None: #hack for dgl saved models
            loaded_dict = loaded_dict['model_state_dict']
        for key, value in loaded_dict.items():
            if key.startswith("module"): state_dict[key[7:]] = value
            else: state_dict[key] = value
        self.load_state_dict(state_dict)

class SpecEncMLP_BIN(nn.Module):
    def __init__(self, params, bin_size):
        super(SpecEncMLP_BIN, self).__init__()

        self.dropout = nn.Dropout(params['fc_dropout'])
        self.mz_fc1 = nn.Linear(bin_size, params['final_embedding_dim'] * 2)
        self.mz_fc2 = nn.Linear(params['final_embedding_dim'] * 2, params['final_embedding_dim'] * 2)
        self.mz_fc3 = nn.Linear(params['final_embedding_dim'] * 2, params['final_embedding_dim'])
        self.relu = nn.ReLU()
        self.aggr_method = params['aggregator']

    def aggr(self, mzvec):
        if self.aggr_method == 'sum':
            aggr_ret = torch.sum(mzvec, axis=1)
        elif self.aggr_method == 'mean':
            aggr_ret = mzvec.sum(axis=1) / ~pad.sum(axis=-1).unsqueeze(-1)
        elif self.aggr_method == 'maxpool':
            input_mask_expanded = torch.where(pad==True, -1e-9, 0.).unsqueeze(-1).expand(mzvec.size()).float()
            aggr_ret = torch.max(mzvec-input_mask_expanded, 1)[0] # Set padding tokens to large negative value

        return aggr_ret

    def forward(self, mzi_b, int_b, pad, lengths):

       h1 = self.mz_fc1(mzi_b)
       h1 = self.relu(h1)
       h1 = self.dropout(h1)
       h1 = self.mz_fc2(h1)
       h1 = self.relu(h1)
       h1 = self.dropout(h1)
       mz_vec = self.mz_fc3(h1)
       mz_vec = self.dropout(mz_vec)

       #mz_vec = self.aggr(mz_vec)

       return mz_vec

    def load_weights(self, pretrained_model):
        # load pretrained_model weights
        state_dict = {}
        loaded_dict = torch.load(pretrained_model, map_location=torch.device('cpu'))
        if loaded_dict.get('model_state_dict', None) is not None: #hack for dgl saved models
            loaded_dict = loaded_dict['model_state_dict']
        for key, value in loaded_dict.items():
            if key.startswith("module"): state_dict[key[7:]] = value
            else: state_dict[key] = value
        self.load_state_dict(state_dict)


class INTER_MLP2(nn.Module):
    def __init__(self, params):
        super(INTER_MLP2, self).__init__()
        self.dropout = nn.Dropout(params['fc_dropout'])
        self.fp_fc1 = nn.Linear(params['final_embedding_dim'] * 2, params['final_embedding_dim'])
        self.fp_fc2 = nn.Linear(params['final_embedding_dim'], params['final_embedding_dim'] // 2)
        self.fp_fc3 = nn.Linear(params['final_embedding_dim'] // 2, params['final_embedding_dim'] // 4)
        self.fp_fc4 = nn.Linear(params['final_embedding_dim'] // 4, params['final_embedding_dim'] // 8)
        self.fp_fc5 = nn.Linear(params['final_embedding_dim'] // 8, 1)
        self.relu = nn.ReLU()
        self.sigmoid = nn.Sigmoid()

    def forward(self, mol, spec):

        y_cat = torch.cat((mol, spec), 1)

        h = self.fp_fc1(y_cat)
        h = self.relu(h)
        h = self.dropout(h)
        h = self.fp_fc2(h)
        h = self.relu(h)
        h = self.dropout(h)
        h = self.fp_fc3(h)
        h = self.relu(h)
        h = self.dropout(h)
        h = self.fp_fc4(h)
        h = self.relu(h)
        h = self.dropout(h)

        h = self.fp_fc5(h)
        z_interaction = self.sigmoid(h)

        return z_interaction

    def load_weights(self, pretrained_model):
        # load pretrained_model weights
        state_dict = {}
        loaded_dict = torch.load(pretrained_model, map_location=torch.device('cpu'))
        if loaded_dict.get('model_state_dict', None) is not None: #hack for dgl saved models
            loaded_dict = loaded_dict['model_state_dict']
        for key, value in loaded_dict.items():
            if key.startswith("module"): state_dict[key[7:]] = value
            else: state_dict[key] = value
        self.load_state_dict(state_dict)