Prediction of circRNA-drug sensitivity using random auto-encoders and multi-layer heterogeneous graph transformers

Prediction of circRNA-drug sensitivity using random auto-encoders and multi-layer heterogeneous graph transformers

연구 분야: Databases

학회: Applied Intelligence

A growing evidence has demonstrated that the expression of circRNAs have significant impact on cell sensitivity to drugs, thereby affecting drug efficacy. Several computational methods have been developed to identify potential circRNA-drug sensitivity associations based on graph auto-encoder and multi-modal information of circRNA and drugs. However, multi-modal information may still lead to a local embedding representation space. And the graph auto-encoder is easy to neglect the global information of the whole graph. Thus, the predictive performance of existing methods is still not satisfactory and needs improvement. In this study, we introduce a model named MHGTCDA for forecasting potential circRNA-drug sensitivity associations using adaptive random auto-encoders (RAEs) and multi-layer Heterogeneous Graph Transformers (MHGT). Firstly, random auto-encoders are used to encode the circRNAs and drugs, respectively. Secondly, MHGT framework is used to obtain context representation of the nodes, which directly utilizes the edge information of the bipartite graph composed of circRNA-drug pairs, thereby reducing information loss. Then, the concatenated embedding matrices of circRNAs and drugs from MHGT are decoded through inner product to obtain the predicted circRNA-drug sensitivity associations. Extensive cross-validation experiments demonstrate that MHGTCDA outperforms nine other state-of-the-art methods. Case studies further illustrate the excellent predictive ability of the proposed method. These results highlight the potential of MHGTCDA as a valuable method for predicting circRNA-drug sensitivity associations, offering significant benefits to drug development.