Abstract
Higher-order genetic interactions, which have profound impact on phenotypic variations, remain poorly characterized. Almost all studies to date have primarily reported pairwise interactions because it is dauntingly difficult to design high-throughput genetic screenings of the large combinatorial search space for higher-order interactions. Here, we develop an algorithm named Dango, based on a self-attention hypergraph neural network, to effectively predict the higher-order genetic interaction for a group of genes. As a proof-of-concept, we make comprehensive prediction of >400 million trigenic interactions in the yeast S. cerevisiae, significantly expanding the quantitative characterization of trigenic interactions. We find that Dango can accurately predict trigenic interactions that reveal both known and new biological functions related to cell growth. The predicted trigenic interactions can also serve as powerful genetic markers to predict growth response to many distinct conditions. Dango enables unveiling a more complete map of complex genetic interactions that impinge upon phenotypic diversity.
Competing Interest Statement
The authors have declared no competing interest.
