Recognizing and Mitigating Reference Bias in Population Genomics.

Most population genomic analyses require mapping sequence reads to a reference genome. Mapping to a single reference introduces reference bias, defined as either low-quality read mapping or downstream analytical distortions arising from biased mapping. Although reference bias has been repeatedly documented, its consequences remain insufficiently understood. Its impacts are particularly relevant in biodiversity genomics, where heterospecific references are commonly used to analyse entire clades. Here, we survey the literature to document where reference bias occurs, its sources, and its impacts, focusing on population and biodiversity genomics. We identify five major classes of reference-inference bias impacts: distortions in population genetic inference; alterations to phylogenetic inference; structural variant detection; RNA expression analyses; and gene presence-absence variation. Critically, the direction and magnitude of these impacts are inconsistent across studies, suggesting that generalisations are premature. Reference bias is primarily driven by evolutionary divergence between the reference genome and the mapped samples, and is further aggravated by analytical and biological factors. We suggest several mitigation strategies, including divergence-aware filtering, tool benchmarking, and standardised reporting of reference bias. In the longer term, we advocate expanding reference genome diversity and adopting pangenome-based alignment across taxa.