1+ # Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
2+ #
3+ # Licensed under the Apache License, Version 2.0 (the "License");
4+ # you may not use this file except in compliance with the License.
5+ # You may obtain a copy of the License at
6+ #
7+ # http://www.apache.org/licenses/LICENSE-2.0
8+ #
9+ # Unless required by applicable law or agreed to in writing, software
10+ # distributed under the License is distributed on an "AS IS" BASIS,
11+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12+ # See the License for the specific language governing permissions and
13+ # limitations under the License.
14+
15+ import paddle
16+ from pgl .utils .data import Dataset , Dataloader
17+ import argparse
18+ import sys
19+ #sys.path.append('.')
20+ from tsnet import TSNet
21+
22+ from utils_no_de import *
23+ from rdkit import Chem
24+ import pandas as pd
25+ import numpy as np
26+
27+ from sklearn .metrics import (roc_auc_score , average_precision_score , f1_score , roc_curve ,
28+ precision_score , recall_score , auc , cohen_kappa_score ,
29+ balanced_accuracy_score , precision_recall_curve , accuracy_score )
30+ from scipy .stats import pearsonr
31+ from sklearn .utils import shuffle
32+
33+ def train (model , data_loader , lincs , loss_fn , opt ):
34+ total_pred , total_lb = [], []
35+ total_loss = []
36+ model .train ()
37+ for g1 , g2 , gm1 , gm2 , cell , lbs in data_loader :
38+ g1 = g1 .tensor ()
39+ g2 = g2 .tensor ()
40+ gm1 = paddle .to_tensor (gm1 , 'int64' )
41+ gm2 = paddle .to_tensor (gm2 , 'int64' )
42+ cell = paddle .to_tensor (cell , 'float32' )
43+ #dea = paddle.to_tensor(dea, 'float32')
44+ #deb = paddle.to_tensor(deb, 'float32')
45+ lbs = paddle .to_tensor (lbs , 'int64' )
46+ #batch_samples = len(lbs)
47+ preds = model (g1 , g2 , gm1 , gm2 , cell , lincs , len (lbs ))
48+ loss = loss_fn (preds , lbs )
49+ loss .backward ()
50+ #print(preds.gradient())
51+ opt .step ()
52+ opt .clear_grad ()
53+ total_loss .append (loss .numpy ())
54+
55+ return np .mean (total_loss )
56+
57+ def eva (model , data_loader , lincs , loss_fn ):
58+ model .eval ()
59+ total_pred , total_lb = [], []
60+ total_loss = []
61+
62+ for g1 , g2 , gm1 , gm2 , cell , lbs in data_loader :
63+ g1 = g1 .tensor ()
64+ g2 = g2 .tensor ()
65+ gm1 = paddle .to_tensor (gm1 , 'int64' )
66+ gm2 = paddle .to_tensor (gm2 , 'int64' )
67+ cell = paddle .to_tensor (cell , 'float32' )
68+
69+ lbs = paddle .to_tensor (lbs , 'int64' )
70+ #batch_samples = len(lbs)
71+ preds = model (g1 , g2 , gm1 , gm2 , cell , lincs , len (lbs ))
72+ loss = loss_fn (preds , lbs )
73+ total_loss .append (loss .numpy ())
74+ total_pred .append (preds .numpy ())
75+ total_lb .append (lbs .numpy ())
76+ total_pred = np .concatenate (total_pred , 0 )
77+ total_lb = np .concatenate (total_lb , 0 )
78+
79+ return total_pred , total_lb , np .mean (total_loss )
80+
81+ def test_auc (model , data_loader , lincs , criterion ):
82+ test_pred , test_label , test_loss = eva (model , data_loader , lincs , criterion )
83+ test_prob = paddle .nn .functional .softmax (paddle .to_tensor (test_pred )).numpy ()[:,1 ]
84+ pred_label = [1 if x > 0.5 else 0 for x in test_prob ]
85+ ACC = accuracy_score (test_label , pred_label )
86+ BACC = balanced_accuracy_score (test_label , pred_label )
87+ PREC = precision_score (test_label , pred_label )
88+ TPR = recall_score (test_label , pred_label )
89+ KAPPA = cohen_kappa_score (test_label , pred_label )
90+
91+ precision , recall , threshold2 = precision_recall_curve (test_label , test_prob )
92+ return roc_auc_score (test_label , test_prob ), auc (recall , precision ), test_loss , ACC , BACC , PREC , TPR , KAPPA
93+
94+
95+
96+ def Pred (model , lincs , data_loader ):
97+ model .eval ()
98+ total_pred = []
99+
100+ for g1 , g2 , gm1 , gm2 , cell , lbs in data_loader :
101+ g1 = g1 .tensor ()
102+ g2 = g2 .tensor ()
103+ gm1 = paddle .to_tensor (gm1 , 'int64' )
104+ gm2 = paddle .to_tensor (gm2 , 'int64' )
105+ cell = paddle .to_tensor (cell , 'float32' )
106+
107+ #lbs = paddle.to_tensor(lbs, 'int64')
108+ #batch_samples = len(lbs)
109+ preds = model (g1 , g2 , gm1 , gm2 , cell , lincs , len (lbs ))
110+
111+ total_pred .append (preds .numpy ())
112+
113+ total_pred = np .concatenate (total_pred , 0 )
114+ total_prob = paddle .nn .functional .softmax (paddle .to_tensor (total_pred )).numpy ()[:,1 ]
115+
116+ return total_prob
117+
118+ def main (args ):
119+ """
120+ Args:
121+ -ddi: drug drug synergy file.
122+ -rna: cell line gene expression file.
123+ -lincs: gene embeddings.
124+ -dropout: dropout rate for transformer blocks.
125+ -epochs: training epochs.
126+ -batch_size
127+ -lr: learning rate.
128+
129+ """
130+ #paddle.set_device('cpu')
131+ ddi = pd .read_csv (args .ddi )
132+ rna = pd .read_csv (args .rna , index_col = 0 )
133+ lincs = pd .read_csv (args .lincs , index_col = 0 , header = None ).values
134+ lincs = paddle .to_tensor (lincs , 'float32' )
135+
136+ ##############independent validation############
137+ #5-fold cross validation
138+ """NUM_CROSS = 5
139+ ddi_shuffle = shuffle(ddi)
140+ data_size = len(ddi)
141+ fold_num = int(data_size / NUM_CROSS)
142+ for fold in range(NUM_CROSS):
143+ ddi_test = ddi_shuffle.iloc[fold*fold_num:fold_num * (fold + 1), :]
144+ ddi_train_before = ddi_shuffle.iloc[:fold*fold_num, :]
145+ ddi_train_after = ddi_shuffle.iloc[fold_num * (fold + 1):, :]
146+ ddi_train = pd.concat([ddi_train_before, ddi_train_after])"""
147+
148+ ddi_train = ddi .copy ()
149+ train_cell = join_cell (ddi_train , rna )
150+ bt_tr = DDsData (ddi_train ['drug1' ].values ,
151+ ddi_train ['drug2' ].values ,
152+ train_cell ,
153+ ddi_train ['label' ].values )
154+
155+ """test_cell = join_cell(ddi_test, rna)
156+ #test_pta, test_ptb = join_pert(ddi_test, drugs_pert)
157+ bt_test = DDsData(ddi_test['drug1'].values,
158+ ddi_test['drug2'].values,
159+ test_cell,
160+
161+ ddi_test['label'].values)"""
162+
163+
164+ loader_tr = Dataloader (bt_tr , batch_size = args .batch_size , num_workers = 4 , collate_fn = collate )
165+ #loader_test = Dataloader(bt_test, batch_size=args.batch_size, num_workers=4, collate_fn=collate)
166+ #loader_val = Dataloader(bt_val, batch_size=args.batch_size, num_workers=1, collate_fn=collate)
167+
168+ model = TSNet (num_drug_feat = 78 ,
169+ num_L_feat = 978 ,
170+ num_cell_feat = rna .shape [1 ],
171+ num_drug_out = 128 ,
172+ coarsed_heads = 4 ,
173+ fined_heads = 4 ,
174+ coarse_hidd = 64 ,
175+ fine_hidd = 64 ,
176+ dropout = args .dropout )
177+ opt = paddle .optimizer .Adam (learning_rate = args .lr , parameters = model .parameters ())
178+ loss_fn = paddle .nn .CrossEntropyLoss ()
179+
180+ for e in range (args .epochs ):
181+ train_loss = train (model , loader_tr , lincs , loss_fn , opt )
182+ print ('Epoch {}---training loss:{}' .format (e , train_loss ))
183+ t_auc , test_prauc , test_loss , acc , bacc , prec , tpr , kappa = test_auc (model , loader_test , lincs , loss_fn )
184+ print ('---Testing loss:{:.4f}, AUC:{:.4f}, PRAUC:{:.4f}, ACC:{:.4f}, BACC:{:.4f}, PREC:{:.4f}, TPR:{:.4f}, KAPPA:{:.4f}'
185+ .format (test_loss , t_auc , test_prauc , acc , bacc , prec , tpr , kappa ))
186+
187+ #paddle.save(model.state_dict(), 'Results/xx.pdparams'.format(e+1))
188+ #model_params = paddle.load('Results/xx.pdparams')
189+ #model.set_state_dict(model_params)
190+
191+
192+ if __name__ == '__main__' :
193+ parser = argparse .ArgumentParser ()
194+ #parser.add_argument("--cuda", action='store_true', default=False)
195+ parser .add_argument ("--dropout" , type = float , default = 0.6 )
196+ parser .add_argument ("--epochs" , type = int , default = 50 )
197+ parser .add_argument ("--batch_size" , type = int , default = 32 )
198+ parser .add_argument ("--lr" , type = float , default = 5e-6 )
199+ parser .add_argument ("--lincs" , type = str , default = '../data/gene_vector.csv' )
200+ parser .add_argument ("--ddi" , type = str , help = 'using SMILES represent drugs' , default = '../data/ddi_dupave.csv' )
201+ parser .add_argument ("--ddi_test" , type = str )
202+ parser .add_argument ("--rna" , type = str , default = '../rna.csv' )
203+
204+ args = parser .parse_args ()
205+ print (args )
206+ main (args )
0 commit comments