Hypergraph contrastive learning with optimal transport: an effective hypergraph contrastive learning framework for deciphering complicated microbe-drug interaction relationships

The complex interplay among microbiota plays a pivotal role in shaping host’s health and disease. Accumulated evidence has demonstrated that microbes are actively involved in modulating drug efficacy and toxicity. As a result, microbiota has emerged as an ideal target for the development of antibacterial agents. Deciphering microbe-drug interaction relationships can significantly accelerate the drug development process. With the increasing availability of pharmacological datasets and microbiome datasets, there is an urgent need to design novel computational approaches to identify latent microbe-drug interactions. Here, we proposed a novel computation framework HCLOT, referring to Hypergraph Contrastive Learning with Optimal Transport. It assumes that there exist high-order interactions between microbes and drugs, and learns the latent representations of both through hypergraph contrastive leaning combined with optimal transport. This approach effectively addresses challenges related to spatial embedding inaccuracy and noise levels. Extensive experiments were implemented on two real datasets. The experimental results show that HCLOT achieves superior performance across multiple evaluation metrics, and provided rich biological insights into microbe-drug interaction mechanisms at the community level. Microbial taxon set enrichment analysis reveals that the enriched taxon sets are directly associated with host-intrinsic factor.