The Raphael Lab at Princeton’s Department of Computer Science and a colleague from the Department of Ecology & Evolutionary Biology found inconsistencies in measuring epistasis, especially in higher-order interactions. Epistasis, or genetic interactions, complicates the fitness landscape, which maps genotype to fitness. The study focuses on the discrepancies between chimeric and multiplicative formulas in quantifying epistasis, particularly in experimental fitness data from haploid genomes:
• The chimeric formula, which measures deviations from a multiplicative fitness model on an additive scale, can yield different magnitudes and signs of epistasis than the multiplicative formula, especially in higher-order interactions (interactions between three or more loci).
• They demonstrate that the chimeric formula isn't appropriate for measuring higher-order epistasis between biallelic mutations.
• Connecting epistasis measures to the multivariate Bernoulli distribution (MVB), they provide a unifying statistical framework, revealing that different epistasis formulas correspond to different parameterizations of the MVB.
• The chimeric epistasis measure corresponds to the joint cumulants of the MVB, which isn't an appropriate measure of higher-order interactions for binary random variables.
• Simulations show that the chimeric formula is less accurate than the multiplicative/additive epistasis formulae and may falsely detect higher-order epistasis.
• Re-analysis of multi-gene knockout data in yeast and multi-way drug interactions in E. coli, as well as deep mutational scanning of several proteins, reveals that a significant percentage of inferred higher-order interactions change sign when using the multiplicative/additive formula compared to the chimeric formula. Approximately 10% to 60% of inferred higher-order interactions change sign using the multiplicative/additive formula compared to the chimeric formula.
• Specifically, in yeast genetics, re-analyzing multi-gene knockout data using the multiplicative formula changes the sign of 12% of trigenic interactions, particularly negative interactions related to functional redundancy. The multiplicative epistasis formula identifies nearly 500 negative interactions not reported using the chimeric formula, extending the trigenic interaction network by 25%.
• In E. coli, the study reanalyzes a drug response dataset, finding that the signs of chimeric and additive interaction measures disagree for multi-way interactions between antibiotics, with the discrepancy increasing with interaction order. The chimeric measure is more likely to identify antagonistic interactions than the additive measure.
The study demonstrates that the chimeric formula can lead to incorrect biological interpretations, especially for higher-order interactions. The authors advocate using mathematically appropriate additive or multiplicative formulas to measure higher-order interactions to obtain more accurate and biologically meaningful results. The appropriate choice of null fitness model depends on the quantitative trait used to approximate fitness.
Chitra U, Arnold B, Raphael BJ. Resolving discrepancies between chimeric and multiplicative measures of higher-order epistasis. Nat Commun. 2025 Feb 17;16(1):1711. doi: 10.1038/s41467-025-56986-5. PMID: 39962081. https://www.nature.com/articles/s41467-025-56986-5
Comentarios