MGTNSyn: Molecular structure-aware graph transformer network with relational attention for drug synergy prediction

Yunjiong Liu, Peiliang Zhang, Dongyang Li, Chao Che, Bo Jin

Abstract: Highlights•We combine atoms’ absolute positions with the chemical information of drugs.•GRAN adaptively identifies and aggregates key local substructures in the drugs.•MGTN is proposed to detect mutagenic motifs and encode the drug’s explicit features.•GRAN and MGTN address the over-smoothing and over-squashing problems in GNN.•Many of the novel drug combinations predicted by MGTNSyn are proven to be effective.