The accurate estimation of the distribution of fitness effects (DFE) of new mutations is critical for population genetic inference but remains a challenging task. While various methods have been developed for DFE inference using the site frequency spectrum of putatively neutral and selected sites, their applicability in species with diverse life history traits and complex demographic scenarios is not well understood. Selfing is common among eukaryotic species and can lead to decreased effective recombination rates in such populations, increasing Hill-Robertson interference between selected mutations. We employ forward simulations to investigate the limitations of current DFE estimation approaches in the presence of selfing and linked effects of selection. We find that distortions of the site frequency spectrum due to Hill-Robertson interference in highly selfing populations lead to an overestimation of the proportion of mildly deleterious mutations. In addition, the proportion of adaptive substitutions estimated at high rates of selfing is overestimated. Our results better clarify the parameter space where current DFE methods might be problematic and where they remain robust in the presence of selfing and other model violations like departures from semi-dominance, population structure, and uneven sampling.
Limitations of the inference of the distribution of fitness effects of new mutations in partially-selfing populations with linkage