GNN_model_v3_50epc_COSINE.py

# ============================ 1. Environment setup ============================
import os
import dgl
import time
import torch
#import pickle
import numpy as np
import pandas as pd
#import networkx as nx
import torch.nn as nn
from tqdm import tqdm
import torch.nn.functional as F
from dgl.data import DGLDataset

# ========================= 2. Setting up the Dataset =========================


data_time = time.time()

class PolypharmacyDataset(DGLDataset):
    def __init__(self):
        super().__init__(name='polypharmacy')

    def process(self):
        features = pd.read_csv('../data/GNN-GSE_full_pkd_norm.csv',index_col = 'ProteinID', sep=',')
        drug_comb = pd.read_csv('../data/GNN-TWOSIDE-train-PSE-964.csv', sep=',') # or 3347
        #properties = pd.read_csv('../data/GNN_properties.csv')
        nodes = pd.read_csv('../data/GNN-GSE_full_pkd_norm.csv', sep=',')
        edges = pd.read_csv('../data/GNN-PPI-net.csv', sep=',')
        dti = pd.read_csv('../data/GNN-DTI_full.csv', sep=',')
        DrugID = pd.read_csv('../data/DrugID.csv', sep = ',')
        print('data loaded!')
        
        # generate drug specific ppi subgraph for GNN edges
        def drug2ppi(drug):
            genes = dti['ProteinID'].loc[dti['DrugID'] == drug].tolist()
            df = edges[['protein1','protein2']].loc[edges['protein1'].isin(genes)]
            df = df.loc[df['protein2'].isin(genes)]
            num_nodes = len(df['protein1'].unique())
            df['graph_id'] = DrugID.loc[DrugID['DrugID'] == drug]['GraphID'].tolist()[0]  #DrugID
            df = df.rename(columns={'protein1': 'src_prot', 'protein2': 'dst_prot'}) # prot: actual protein id
            final_genes =df['src_prot'].unique().tolist() # final genes that have ppi data
            dic = {gene:final_genes.index(gene) for gene in final_genes} # conversion dic, starts at 0
            df['src'] = df['src_prot'].map(dic) #local ids
            df['dst'] = df['dst_prot'].map(dic) #local ids
            return(df[['graph_id', 'src', 'dst', 'src_prot', 'dst_prot']],num_nodes)
        
        self.graphs = []
        self.labels = []
        self.comb_graphs = []
        self.comb_labels = []

        #Node features or PSEs dictionary
        feature_dic = {i+1:torch.tensor(features.loc[i+1,]) for i in range(len(features))}
    
        # For each graph ID...
        for drug in tqdm(DrugID['DrugID'].tolist()):
            # Find the edges as well as the number of nodes and its label.
            edges_of_id,num_nodes = drug2ppi(drug)
            src = edges_of_id['src'].to_numpy()
            dst = edges_of_id['dst'].to_numpy()
            label = DrugID.loc[DrugID['DrugID'] == drug]['DrugID'].tolist()[0]
            
            # Create a graph and add it to the list of graphs and labels.
            g = dgl.graph((src, dst), num_nodes=num_nodes)
            
            # Need to convert proteinsIDs for feature assigning
            prot_ids = edges_of_id['src_prot'].unique().tolist()
            for prot in edges_of_id['dst_prot'].unique().tolist():
                if prot not in prot_ids:
                    prot_ids.append(prot)
            convert_prot = {prot_ids.index(prot):prot for prot in prot_ids}
            
            #Adding features of each node
            g.ndata['PSE'] = torch.zeros(g.num_nodes(), 964)
            for node in g.nodes().tolist():
                g.ndata['PSE'][node] = feature_dic[convert_prot[node]]
                
            self.graphs.append(g)
            self.labels.append(label)
 
        # conver drugid to their respective graph id
        #drug2graph = {properties['label'][i]:i for i in range(len(properties))} 
        #drug2graph = {self.labels[i]:i for i in range(len(self.labels))} 
        
        for i in range(len(drug_comb)):
            row = drug_comb.loc[i]
            g1 = self.graphs[self.labels.index(row[0])] # Drug1 graph
            g2 = self.graphs[self.labels.index(row[1])] # Drug2 graph 
            self.comb_graphs.append([g1,g2])
            self.comb_labels.append(torch.tensor(row[2:])) # PSE values
   

    def __getitem__(self, i):
        return self.comb_graphs[i], self.comb_labels[i]
        #return self.graphs[i], self.labels[i]

    def __len__(self):
        return len(self.comb_graphs)

    
print('\nCreating the PolypharmacyDataset ...\n')

dataset = PolypharmacyDataset()

print('\nPolypharmacyDataset created!\n')

end = time.time()
hours, rem = divmod(end-data_time , 3600)
minutes, seconds = divmod(rem, 60)

print('\ndataset is compiled! \ncompiling time = {:0>2}:{:0>2}:{:05.2f}'.format(int(hours),int(minutes),seconds))

# ========================= 3. Data loading and batch =========================

print('\nCreating train and test batches ...\n')

from dgl.dataloading import GraphDataLoader
from torch.utils.data.sampler import SubsetRandomSampler

num_examples = len(dataset)
num_train = int(num_examples * 0.05)

train_sampler = SubsetRandomSampler(torch.arange(num_train))
#test_sampler = SubsetRandomSampler(torch.arange(num_train, num_examples))

train_dataloader = GraphDataLoader( dataset, sampler=train_sampler, batch_size=35 ,drop_last=False)
#test_dataloader = GraphDataLoader( dataset, sampler=test_sampler, batch_size=35 ,drop_last=False)

print('\nTrain and test batches are created!\n')

# ========================= 4. GNN Model: Siamese GCN =========================

from dgl.nn import GraphConv

class GCN(nn.Module):
    def __init__(self, in_feats, h_feats, num_classes):
        super(GCN, self).__init__()
        self.conv1 = GraphConv(in_feats, h_feats)
        self.conv2 = GraphConv(h_feats,  num_classes)
        
    def forward(self, g, in_feat):
        h = self.conv1(g, in_feat)
        h = F.relu(h)
        h = self.conv2(g, h)
        g.ndata['h'] = h
        out = F.relu(dgl.mean_nodes(g, 'h'))
        #out = F.relu(dgl.max_nodes(g, 'h'))
        return out

# =============================== 5. Training =================================
train_time = time.time()

#print('\nCreating the SiameseGCN model ...\n')

# Create the model with given dimensions
model = GCN(964, 200, 964)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

print('\n======================== Trainig ========================\n')

f = open("log_model_V3_Cosine.txt", "a")
f.write('\n======================== Trainig ========================\n')

for epoch in range(50):
    batch = 0
    for batched_graph, labels in train_dataloader:
        start_time = time.time()
        g1 = batched_graph[0]
        g2 = batched_graph[1]
        pred1 = model(g1, g1.ndata['PSE'].float())
        pred2 = model(g2, g2.ndata['PSE'].float())
        pred = F.normalize(pred1+pred2)/2
        #loss = F.binary_cross_entropy(torch.sigmoid(pred).float(),labels.float())
        loss = 1- F.cosine_similarity(torch.sigmoid(pred),labels).mean()
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        run_time = round(time.time() - start_time,3)
        msg = 'epoch%s, batch%s | loss = %s | time: %s s\n' % (epoch,batch,loss.tolist(),run_time)
        f.write(msg)
        print (msg)
        batch += 1

        TP = 0
        FP = 0
        TN = 0
        FN = 0
        for i in range(len(labels)):
            for j in range(len(labels[i])):
                if labels[i][j] == 1 and pred[i][j] != 0:
                    TP += 1
                elif labels[i][j] == 1 and pred[i][j] == 0:
                    FN += 1
                elif labels[i][j] == 0 and pred[i][j] != 0:
                    FP += 1
                elif labels[i][j] == 0 and pred[i][j] == 0:
                    TN += 1
                else:
                    pass
            
        # Validation metrics        
        acc = ((TP+TN)*100)/(TP+FP+FN+TN)
        prec = (TP*100)/(TP+FP)
        recall = (TP*100)/(TP+FN)
        F1 = 2*(recall*prec)/(recall+prec)
        sim = ((F.cosine_similarity(pred.float(),labels.float())).mean().tolist())*100
        msg2 = 'Accuracy: %s | Precision: %s | Recall: %s | F1: %s | Similarity: %s\n' %(round(acc,4),round(prec,4),round(recall,4),round(F1,4),round(sim,4))
        f.write(msg2)
        print(msg2)



end = time.time()
hours, rem = divmod(end-train_time , 3600)
minutes, seconds = divmod(rem, 60)

msg_tiem = '\ntotal training time = {:0>2}:{:0>2}:{:05.2f}'.format(int(hours),int(minutes),seconds)
f.write(msg_tiem)
print(msg_tiem)

torch.save(model.state_dict(), 'state_dict_model_V3_cosine.pt')
torch.save(model, 'entire_model_V3_cosine.pt')
print('\nmodel is saved!\n')
f.write('\nmodel saved!\n')


f.close()