MAF-CDR: Multi-Omics Data Integration for Cancer-Drug Response Prediction Model
Sizhe Zhang, Xuecong Tian, Su Ying, Wanhua Huang, Xiaoyi Lv, Keao Li · 2023
Achieving precise application and rapid discovery of anticancer drugs has been an important topic in medicine and pharmacology. Traditional anticancer drug discovery relies on in vivo experiments and in vitro drug screening, methods that are important in the discovery of new drugs, yet require more time and resources. Although there are some traditional methods based on machine learning to predict drug-cancer cell lineage response (CDR), these methods are usually based on a single source of information, limiting the interpretation of the panoramic view of cancer cells. Using multi-omics information to comprehensively respond to cancer cells and anticancer drug responses remains a great challenge. In this experiment, we propose MAF-CDR, a CDR prediction model based on the alignment and fusion of multi-omics information. The model fuses multi-omics information in a unified framework and achieves cross-modal information representation through adversarial training to achieve aligned fusion of different modalities. Meanwhile, in order to better capture the correlation between drugs and cancer cell lines, MAF-CDR constructed a graph neural network encoder based on contrast learning to achieve accurate CDR prediction. The experimental results showed that the AUC value of MAF-CDR was higher than 0.8, which was better than other current baseline methods. This indicates that the MAF-CDR method can efficiently promote the development and innovation in the field of drug discovery, and proves the great potential value of deep learning methods in guiding the rapid discovery and precise application of anti-cancer drugs.