Dango: Predicting higher-order genetic interactions

Area of Science:

• Genetics
• Computational Biology
• Systems Biology

Background:

• Higher-order genetic interactions are crucial for phenotypic variation but are poorly understood due to screening challenges.
• Current research primarily focuses on pairwise interactions, leaving complex genetic relationships under-characterized.

Purpose of the Study:

• To develop a computational method for predicting higher-order genetic interactions among groups of genes.
• To expand the quantitative landscape of trigenic interactions and their biological implications.

Main Methods:

• Developed Dango, a self-attention hypergraph neural network.
• Applied Dango to predict over 400 million trigenic interactions in yeast (S. cerevisiae).
• Incorporated protein embeddings and uncertainty estimation for enhanced biological relevance and interpretability.

Main Results:

• Dango accurately predicts trigenic interactions, revealing functions related to cell growth.
• Predicted interactions serve as genetic markers for growth responses under various conditions.
• Significantly expanded the known quantitative landscape of trigenic interactions.

Conclusions:

• Dango enables a more comprehensive mapping of complex genetic interactions.
• This method facilitates a deeper understanding of the molecular mechanisms underlying phenotypic diversity.
• The findings highlight the importance of higher-order genetic interactions in biological systems.