mldlproject/2024-iACP-GCR
Source code and data of the paper entitled "iACP-GCR: Identifying multi-target anticancer compounds using multitask learning on graph convolutional residual neural networks"
iACP-GCR: Multitask Learning on Graph Convolutional Residual Neural Networks for Screening of Multi-target Anticancer Compounds
T.-H Nguyen-Vo, Trang T. T. Do, and B. P. Nguyen∗
Motivation
The finding of promising anticancer compounds is vital in anticancer drug development. Recently, besides experimental
approaches, several computational advances have been developed to assist experimental scientists to identify potential
anticancer candidates. In this study, we proposed a computational framework called iGCR-ACP to predict anticancer
compounds. The iGCR-ACP was developed using the NCI-60 dataset, one of the most reliable and well-known sources of experimentally
verified compounds. From the dataset, compounds were screened across nine cancer types (panels), including breast, central
nervous system, colon, leukemia, non-small cell lung, melanoma, ovarian, prostate, and renal, were collected.
Results
The area under the receiver operating characteristic (ROC) curve (AUCROC) values for all prediction tasks were over 0.95. For the area
under the precision-recall (PR) curve (AUCPR) values, the prediction task for the breast panel obtained the smallest AUCPR value of about 0.76
while those for the leukemia and colon panels achieved the greatest AUCPR values of about 0.80. The results showed that the iACP-GCR
is a robust and stable framework for determining compounds possessing anticancerous activities. Also, the comparative analysis
demonstrated that our approach had achieved better performance compared to state-of-the-art methods.
Availability and implementation
Source code and data are available upon request.