MTDS: Meta-Path Context Enhanced Drug Combination Synergy Prediction

Xiaoyi Liang, Xiaowen Wang, Hongming Zhu, Qin Liu

Abstract: Drug combinations enhance cancer treatment efficacy and reduce toxicity. However, the vast number of potential combinations makes traditional clinical trials impractical. In the context of biological heterogeneous network, existing graph neural network based methods lack attention to meaningful biological relationships. Meanwhile, metapath based approaches often rely on expert knowledge to predefine metapaths, neglecting the importance of intermediate nodes within biological networks. To address this issue, we proposed a multi-task method based on metapath contexts (MTDS) to improve synergistic drug combinations prediction. MTDS automatically constructs metapath contexts of biological heterogeneous graph and adaptively learns node representations. We evaluated MTDS on O’Neil and NCI-ALMANAC datasets, and the results show that MTDS surpasses four state-of-the-art methods. Specifically, MTDS achieved MSE of 222.7143 and PCC of 0.7644 on the O’Neil dataset, and MSE of 84.4580 and PCC of 0.7773 on the NCI-ALMANAC dataset. The experiments indicate that metapath contexts can retain more semantic information from biological heterogeneous graphs, thereby enhancing prediction accuracy.

External IDs:doi:10.1007/978-981-96-7033-8_26