Accurate runs of homozygosity estimation from low coverage genome sequences in non-model species

Runs of homozygosity (ROH) are increasingly being analyzed using whole genome sequences in non-model species as a measure of inbreeding and to assess demographic history, thus providing useful information for conservation. However, most studies have used Plink for ROH inference which performs poorly when sequencing depth is below 10X, often underestimating ROH. This can lead to erroneous status assessment and poor management decisions. We assessed the performance of ROHan, a program developed for ROH and heterozygosity estimation using lower coverage sequences that have so far only optimized for human data. Using high coverage whole genomes from 22 caribou, a non-model species at risk presenting varying levels of inbreeding, we assessed the effects of sequencing depth (1X to 15X), the input parameter 'rohmu' that determines the heterozygosity rate which is tolerated within ROH regions, and demographic history on the ROH inference and heterozygosity. Accurate estimation of the percentage of the genome and lengths of ROH could be achieved at depths as low as 3-5X. However, the rohmu parameter and individual demographic history had a significant effect on the results. Heterozygosity was also overestimated at low depth. Using our optimized rohmu parameter, we re-analyzed low coverage sequences from a small and isolated caribou population and demonstrated high inbreeding levels that had previously been missed. We provide recommendations for optimization of the rohmu parameter and demonstrate the need for careful interpretation of outputs to enable robust ROH inference using low coverage whole genome sequences in wildlife species.