Over the past few decades, the rapid democratization of high-throughput sequencing and the growing emphasis on open science practices have resulted in an explosion in the amount of publicly available sequencing data. This opens new opportunities for combining datasets to achieve unprecedented sample sizes, spatial coverage, or temporal replication in population genomic studies. However, a common concern is that non-biological differences between datasets may generate batch effects that can confound real biological patterns. Despite general awareness about the risk of batch effects, few studies have examined empirically how they manifest in real datasets, and it remains unclear what factors cause batch effects and how to best detect and mitigate their impact bioinformatically. In this paper, we compare two batches of low-coverage whole genome sequencing (lcWGS) data generated from the same populations of Atlantic cod (Gadus morhua). First, we show that with a “batch-effect-naive” bioinformatic pipeline, batch effects severely biased our genetic diversity estimates, population structure inference, and selection scan. We then demonstrate that these batch effects resulted from multiple technical differences between our datasets, including the sequencing instrument model/chemistry, read type, read length, DNA degradation level, and sequencing depth, but their impact can be detected and substantially mitigated with simple bioinformatic approaches. We conclude that combining datasets remains a powerful approach as long as batch effects are explicitly accounted for. We focus on lcWGS data in this paper, which may be particularly vulnerable to certain causes of batch effects, but many of our conclusions also apply to other sequencing strategies.

Low-coverage whole genome sequencing (lcWGS) has emerged as a powerful and cost-effective approach for population genomic studies in both model and non-model species. However, with read depths too low to confidently call individual genotypes, lcWGS requires specialized analysis tools that explicitly account for genotype uncertainty. A growing number of such tools have become available, but it can be difficult to get an overview of what types of analyses can be performed reliably with lcWGS data and how the distribution of sequencing effort between the number of samples analyzed and per-sample sequencing depths affects inference accuracy. In this introductory guide to lcWGS, we first illustrate that the per-sample cost for lcWGS is now comparable to RAD-seq and Pool-seq in many systems. We then provide an overview of software packages that explicitly account for genotype uncertainty in different types of population genomic inference. Next, we use both simulated and empirical data to assess the accuracy of allele frequency estimation, detection of population structure, and selection scans under different sequencing strategies. Our results show that spreading a given amount of sequencing effort across more samples with lower depth per sample consistently improves the accuracy of most types of inference compared to sequencing fewer samples each at higher depth. Finally, we assess the potential for using imputation to bolster inference from lcWGS data in non-model species, and discuss current limitations and future perspectives for lcWGS-based analysis. With this overview, we hope to make lcWGS more approachable and stimulate broader adoption.