A multi-view multi-omics model for cancer drug response prediction

Cancer drug response prediction is the fundamental task in precision medicine, which provides opportunities for cancer therapy. Several methods have been proposed to screen drugs, via building computational models on multi-omics data. However, the view value missing problem caused by unknown cancers or tumors has not been addressed. For this reason, a multi-view multi-omics (MvMo) model is proposed to predict cancer drug response values. The proposed MvMo model first represents the input heterogeneous data in different kinds of embeddings and features, such as token embeddings and latent features. Then several views are generated to observe interconnections among those representations. Finally, the predictions are generated based on the outputs of these views. Experimental results on the collected real data show the efficiency of the proposed method in terms of speed and accuracy.