Selection Estimation from Genetic Time-Series Data: Effects of Limited Sampling and Genetic Drift

Estimating selection from genetic time-series data is fundamental to understanding evolutionary dynamics. Accurate selection inference is confounded by multiple noise sources, including limited sampling of populations and genetic drift. To characterize how these uncertainties collectively affect estimator performance, we analyze a mathematically tractable selection coefficient estimator derived under the marginal path likelihood (MPL) framework. We identify a parameter, the integrated mutant allele variance, as a key quantity determining estimator precision. Our analysis reveals that variance integration mitigates sampling and genetic drift errors at different rates, with drift typically becoming the dominant source of error in longer trajectories. The increased robustness of MPL-based estimation to sampling is surprising, since MPL is derived from a model that neglects this effect. Our findings offer insights into how incorporating temporal information reduces multiple sources of noise when estimating selection coefficients.