Drug-combination-synergy/DrugCombPred_O'Neil/predictor/model/models.py

import torch
import torch.nn as nn
import torch.nn.functional as F
import os
import sys 
import pandas as pd
import time
PROJ_DIR = os.path.dirname(os.path.abspath(os.path.join(os.path.dirname( __file__ ), '..')))

sys.path.insert(0, PROJ_DIR)
from drug.models import GCN
from drug.datasets import DDInteractionDataset
from model.utils import get_FP_by_negative_index, get_FP_by_negative_indices
from const import Drug2FP_FILE



class Connector(nn.Module):
    def __init__(self, gpu_id=None):
        self.gpu_id = gpu_id
        super(Connector, self).__init__()
        # self.ddiDataset = DDInteractionDataset(gpu_id = gpu_id)
        self.gcn = None
        self.drug2FP_df = pd.read_csv(Drug2FP_FILE)
        
        #Cell line features
        # np.load('cell_feat.npy')

    def forward(self, drug1_idx, drug2_idx, cell_feat, subgraph):
        if self.gcn == None:
            # print("here is for initializing the GCN. num_features: ", subgraph.num_features)
            self.gcn = GCN(subgraph.num_features, subgraph.num_features // 2)
        # print("this is subgraph in connector model forward: --------------")
        # print(subgraph)
        # graph.get().x --> DDInteractionDataset
        # subgraph = graph.get() --> Data 
        x = subgraph.x
        edge_index = subgraph.edge_index
        x = self.gcn(x, edge_index)
        
        # print("node local indices:")
        node_indices = edge_index.flatten().unique()
        # print(node_indices)

        # print("-----------------------")
        # print("node global indices:")
        node_indices = subgraph.n_id
        if self.gpu_id is not None:
            node_indices = node_indices.cuda(self.gpu_id)
        # print(node_indices)
        
        drug1_idx = torch.flatten(drug1_idx)
        drug2_idx = torch.flatten(drug2_idx)
        #drug1_feat = x[drug1_idx]
        #drug2_feat = x[drug2_idx]
        drug1_feat = torch.empty((len(drug1_idx), len(x[0])))
        drug2_feat = torch.empty((len(drug2_idx), len(x[0])))

        print("x shape: ", x.size())
        print("node_indices: ", node_indices.size())

        start_time = time.time()
        # for index, element in enumerate(drug1_idx):
        #     x_element = element
        #     if element >= 0:
        #         x_element = (node_indices == element).nonzero().squeeze()
        #         drug1_feat[index] = (x[x_element])
        # for index, element in enumerate(drug2_idx):
        #     x_element = element
        #     if element >= 0:
        #         x_element = (node_indices == element).nonzero().squeeze()
        #         drug2_feat[index] = (x[x_element])

        mask_positive = (drug1_idx >= 0)
        x_elements_positive = (node_indices.unsqueeze(-1) == drug1_idx[mask_positive]).nonzero(as_tuple=True)[0]
        drug1_feat[mask_positive] = x[x_elements_positive]

        mask_negative = ~mask_positive
        drug1_feat[mask_negative] = get_FP_by_negative_indices(drug1_idx[mask_negative], self.drug2FP_df)

        mask_positive = (drug2_idx >= 0)
        x_elements_positive = (node_indices.unsqueeze(-1) == drug2_idx[mask_positive]).nonzero(as_tuple=True)[0]
        drug2_feat[mask_positive] = x[x_elements_positive]
        
        if self.gpu_id is not None:
            drug1_feat = drug1_feat.cuda(self.gpu_id)
            drug2_feat = drug2_feat.cuda(self.gpu_id)

        print("first: ", time.time() - start_time)
        start_time = time.time()

        for i, x in enumerate(drug1_idx):
            if x < 0:
                drug1_feat[i] = get_FP_by_negative_index(x,self.drug2FP_df)
        for i, x in enumerate(drug2_idx):
            if x < 0:
                drug2_feat[i] = get_FP_by_negative_index(x,self.drug2FP_df)
        print("second: ", time.time() - start_time)
        
        feat = torch.cat([drug1_feat, drug2_feat, cell_feat], 1)
        return feat


class MLP(nn.Module):
    def __init__(self, input_size: int, hidden_size: int, gpu_id=None):
        super(MLP, self).__init__()
        self.layers = nn.Sequential(
            nn.Linear(input_size, hidden_size),
            nn.ReLU(),
            nn.BatchNorm1d(hidden_size),
            nn.Linear(hidden_size, hidden_size // 2),
            nn.ReLU(),
            nn.BatchNorm1d(hidden_size // 2),
            nn.Linear(hidden_size // 2, 1)
        )

        self.connector = Connector(gpu_id)
    
    # prev input: self, drug1_feat: torch.Tensor, drug2_feat: torch.Tensor, cell_feat: torch.Tensor, subgraph: related subgraph for the batch
    def forward(self, drug1_idx, drug2_idx, cell_feat, subgraph):
        start = time.time()
        feat = self.connector(drug1_idx, drug2_idx, cell_feat, subgraph)
        print("Connector forward time: ", time.time() - start)
        out = self.layers(feat)
        return out


# other PRODeepSyn models have been deleted for now