Authors:

Yucheng Jin

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Genetic interactions occur when mutations in different genes combine to produce unexpected effects on cellular fitness, revealing how genes cooperate within biological pathways and informing precision therapeutic strategies [1,2]. Despite large-scale mapping efforts in Saccharomyces cerevisiae, no consensus exists regarding the optimal mathematical definition for quantifying interaction scores from mutant fitness data. Current approaches rely on baseline models developed across evolutionary biology and population genetics, yet these definitions rest on distinct and sometimes conflicting assumptions about how independent mutations combine. This project evaluates five quantitative models—Min, Product, Log, Additive, and lsing—for both pairwise and triple-gene interactions using high-throughput yeast datasets [3]. The first four models have been widely employed in biological contexts: the Product and Log models assume multiplicative independence, the Additive model assumes linear effects, and the Min model assumes that the more severe mutation masks the other. In contrast, we newly prapose the Ising model, originally developed to describe interacting spins in disordered magnetic systems (spin glasses) in statistical physics, as a framework for capturing gene-gene coupling without assuming strict independence or masking. Each digenic model was generalized into link form and extended to trigenic interactions. Model performance was assessed statistically by examining variance, bias, and residual error between observed and expected mutant fitness. Biological relevance was evaluated using precision—recall curves comparing interaction scores to protein-protein interaction networks and Gene Ontology co-annotation data. For pairwise interactions, the Product and Log models showed strong biological enrichment, while the Min model performed poorly. Notably, in triple-gene analyses, the Ising model produced interaction scores with reduced variance and bias while maintaining low residual error and strong predictive performance. These findings suggest that higher-order mutant data provide critical discriminatory power and support the Ising framework as a promising physics-inspired approach for modeling complex genetic interactions.

Source:

Northwestern University

Topics:

No topics listed

Co-authors:

Yucheng Jin