How precise are mutation rate estimates? Comparison of different approaches to estimate de novo mutation rates

Availability of de novo mutation rate (µ) estimates based on approaches that rely on bioinformatic validations has increased tremendously during the past few years, but the accuracy and precision of these estimates often remain unclear as Sanger sequencing validation of the mutations is often lacking. We used both long- and short-read sequencing data and different bioinformatic pipelines to estimate µ, as well as false positive (FPR) and negative (FNR) rates, for family trios of flat-headed loaches (Oreonectes platycephalus). By comparing estimates against PCR-verified mutations, we observed that the top-performing approach (as ranked by the F1 score of seven approaches at the same depth) still exhibited a 4% false positive rate (FPR) alongside a 12% false-negative rate (FNR). Across the remaining methods, FPR values ranged from 4-12%, and FNRs from 8-19%. Irrespective of the bioinformatic approach used, long-read data yielded consistently lower µ estimates than short-read data because of the larger callable genome sizes. In addition, a higher mapping depth resulted in a lower FNR. These results call for caution regarding de novo mutations without Sanger sequencing validation in non-model organisms and raise the possibility that many published µ-estimates, especially those based on low mapping depths, might be biased.