Mutation choice biases the structure of empirical fitness landscapes
- 1National Institute for Theory and Mathematics in Biology
Abstract
Evolution can be thought of as taking place in a fitness landscape which maps each genotype to a fitness value. This abstraction has motivated efforts to quantify the statistics of random fitness landscapes with different amounts and types of interactions among loci, or epistasis. High-throughput experiments measuring the phenotypes of thousands of genotypes promise to bridge this theory with data. Yet thousands is still small relative to genotype space. Moreover, experiments often assay a highly non-random subset of genotypes. For example, a common approach is to assay all 2^M intermediates between ancestral and evolved sequences differing by M much fewer than L mutations. Here, we ask how well the statistics of a landscape constructed under distinct ascertainment protocols recapitulate those of a random subset of the global landscape in several random landscape models. We show that, in most landscapes, selecting mutations that occurred along an adaptive trajectory inflates the variance explained by lower-order terms, reduces the number of local maxima, and increases the number of accessible paths relative to a random fitness landscape. Selecting the most beneficial or deleterious mutations instead distorts these statistics in ways that depend more critically on the amount of epistasis in the global landscape. We then use three large-scale mutagenesis datasets to simulate ascertainment in more realistic settings. In two of these, differences among schemes are likely explained by measurement noise and global nonlinearities; in the third, these differences at least partly reflect biased sampling of genuine interactions among loci.
Related
- From single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
- Representation learning of somatic missense mutation landscapes across cancer types
- A diffusion model of viral evolution predicts mutation fitness and evolutionary trajectories
- Distinct genomic landscapes of mismatch-repair deficiency (dMMR)/microsatellite instability-high (MSI-H) gastrointestinal (GI) cancer depending on tumor mutation burden (TMB) levels.
- Sequence-structured fitness landscapes predict influenza evolution
BibTeX
@misc{brooks2026mutation,
title = {{Mutation choice biases the structure of empirical fitness landscapes}},
author = {C. Brooks and M. M. Desai and M. O. Carlson},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.28.754994},
url = {https://www.biorxiv.org/content/10.64898/2026.09.28.754994v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Mutation choice biases the structure of empirical fitness landscapes
AU - C. Brooks
AU - M. M. Desai
AU - M. O. Carlson
PY - 2026
DA - 2026/09/28
PB - bioRxiv
DO - 10.64898/2026.09.28.754994
UR - https://www.biorxiv.org/content/10.64898/2026.09.28.754994v1
AB - Evolution can be thought of as taking place in a fitness landscape which maps each genotype to a fitness value. This abstraction has motivated efforts to quantify the statistics of random fitness landscapes with different amounts and types of interactions among loci, or epistasis. High-throughput experiments measuring the phenotypes of thousands of genotypes promise to bridge this theory with data. Yet thousands is still small relative to genotype space. Moreover, experiments often assay a highly non-random subset of genotypes. For example, a common approach is to assay all 2^M intermediates between ancestral and evolved sequences differing by M much fewer than L mutations. Here, we ask how well the statistics of a landscape constructed under distinct ascertainment protocols recapitulate those of a random subset of the global landscape in several random landscape models. We show that, in most landscapes, selecting mutations that occurred along an adaptive trajectory inflates the variance explained by lower-order terms, reduces the number of local maxima, and increases the number of accessible paths relative to a random fitness landscape. Selecting the most beneficial or deleterious mutations instead distorts these statistics in ways that depend more critically on the amount of epistasis in the global landscape. We then use three large-scale mutagenesis datasets to simulate ascertainment in more realistic settings. In two of these, differences among schemes are likely explained by measurement noise and global nonlinearities; in the third, these differences at least partly reflect biased sampling of genuine interactions among loci.
ER -