New Tool Reads Genetic Diversity from Low-Coverage Genome Skims Without a Reference

Tracking the genetic health of populations has long demanded a difficult trade-off. Whole-genome sequencing delivers the richest picture of diversity within and between populations, but it remains expensive enough that most of Earth’s species have never been sequenced at high coverage, let alone repeatedly sampled across seasons, habitats, and generations. A cheaper alternative, known as genome skimming, sequences genomes at very low coverage, generating short reads that scatter thinly across the genome. The problem has always been analysis: at such shallow coverage, assembling a genome is impossible and correcting sequencing errors is unreliable. A new computational method called DipSkmer, described in PLOS Computational Biology by Eduardo Charvel, Homère J. Alves Monteiro, Siavash Mirarab, and Vineet Bafna, now shows that those skimpy datasets can be turned into meaningful estimates of population genetic diversity, even without any reference genome at all.

The quantity at the heart of the method is the population-scaled mutation parameter, written as θ, which ecologists and conservation biologists treat as an essential biodiversity variable for monitoring population health. θ captures the amount of genetic variation expected within a population and underpins many downstream inferences about effective population size, demographic history, and resilience to environmental change. Traditionally, estimating θ required either high-quality whole genomes or aligned sequencing reads mapped onto a reference. Both requirements are stumbling blocks for the vast majority of organisms, particularly rare, non-model, or poorly characterized species that are often the most urgent conservation priorities. DipSkmer removes the reference entirely, working directly from the raw k-mers, the short words of DNA of fixed length, that appear in sequencing reads.

To understand what DipSkmer does differently, it helps to look at the tools it builds upon. Existing alignment-free methods such as Mash estimate the genomic distance between two samples by comparing their sets of k-mers using sketching techniques, which sample a small subset of k-mers to keep computations fast. The comparison relies on the Jaccard index, the fraction of k-mers shared between two sets relative to those found in either set. A later method, Skmer, extended this framework by modeling the effects of low sequencing coverage, allowing it to correct for the fact that shallow data miss many k-mers altogether. These approaches have proven successful for species identification and phylogenetic reconstruction, where the differences between genomes are large. But at the population level, where genomes differ by only a fraction of a percent, they run into a fundamental problem.

That problem is diploidy. Most animals and plants carry two copies of their genome, one inherited from each parent, and those copies differ from each other at heterozygous sites. Mash and Skmer implicitly treat each genome as a single haploid sequence, so the k-mer set of a diploid individual is a mixture of two haplotypes. Heterozygosity therefore inflates the apparent distance between two diploid samples, confounding the very signal that population geneticists want to measure. Two individuals from the same population can look artificially divergent simply because each carries its own complement of heterozygous variants. For population genomics, where the distances of interest are often below two percent, this confounding is not a minor nuisance; it can swamp the signal entirely.

DipSkmer’s central advance is a mathematical treatment of exactly this issue. The authors turned to coalescent theory, the branch of population genetics that describes how genealogies of sampled chromosomes behave under neutrality, and worked under Kingman’s coalescent model to derive how the Jaccard index between the k-mer sets of two diploid samples changes as a function of θ. Instead of treating shared k-mers as a simple measure of genome similarity, the derivation accounts for the fact that some shared k-mers reflect recent shared ancestry between homologous chromosomes, while others arise from heterozygous variation within individuals. The result is an explicit formula linking the observable Jaccard index to the underlying population mutation parameter, turning a heuristic similarity score into a statistically grounded estimator.

Deriving the formula was only the first step, because real sequencing data depart from idealized models in two important ways. First, low coverage means many genomic positions are never observed, so k-mers that exist in the genome are missing from the read data. Second, sequencing errors corrupt the reads, injecting spurious k-mers that are not present in any genome. Charvel and colleagues extended their coalescent-based framework to model both effects, deriving an estimator of θ from the observed Jaccard index alongside estimates of several auxiliary quantities, including the sequencing coverage and the error rate, which are themselves inferred from the skim data. This joint estimation matters because coverage and error distort the Jaccard index in opposite directions: missing coverage reduces the apparent sharing of k-mers, while errors reduce it further by breaking up true k-mers and adding false ones. Disentangling the three effects is what allows DipSkmer to recover an accurate diversity estimate from deliberately shallow data.

The practical payoff is a pipeline that requires nothing more than the skim itself. No reference genome, no alignment step, and no assembly are needed, which means the method can be applied to organisms for which no genomic resources exist. Because genome skimming is dramatically cheaper than high-coverage sequencing, the approach opens the door to the kind of repeated, spatially and temporally extensive sampling that biodiversity monitoring actually requires. A conservation program could, in principle, survey genetic diversity across dozens of populations every year, using the same low-cost sequencing that is already routine for cataloging species in bulk samples.

How well does it work? The authors evaluated DipSkmer on both simulated and empirical datasets, comparing it against the established alignment-free methods Mash and Skmer and against ANGSD, a widely used tool that estimates population genetic parameters from aligned reads. The comparisons show that for diploid organisms and low levels of divergence, specifically distances below about two percent, DipSkmer’s estimates are more accurate than those of the existing alignment-free methods, which struggle precisely in the regime that matters for population genetics. Just as importantly, DipSkmer’s estimates closely approximate those produced by ANGSD, which relies on a reference genome and alignment. In other words, the reference-free approach achieves accuracy comparable to alignment-based analysis in the low-diversity setting, while remaining applicable to species that lack references altogether.

The significance of this result extends beyond a single methodological improvement. Genetic diversity is one of the essential biodiversity variables that monitoring frameworks use to track the state of life on Earth, yet it remains among the hardest to measure at scale. High-quality assembled genomes will not represent the full diversity of life for a considerable time, and even when they do, ongoing monitoring will demand repeated sampling across time and space, something only affordable methods can sustain. By making θ estimable from the cheapest form of genome sequencing available, DipSkmer aligns the computational toolkit with the practical constraints of field biology. The method also inherits the scalability of sketching techniques, since it operates on compact k-mer summaries rather than full read sets, keeping memory and computation manageable even for large collections of samples.

There are, of course, boundaries to what the method can do. Its derivation assumes a neutral Kingman’s coalescent, so populations with strong selection, extreme demographic histories, or unusual reproductive systems may deviate from the model’s assumptions, and the accuracy advantage is demonstrated for diploid samples at low distances rather than for deeply divergent lineages, where the older tools remain appropriate. Still, the trajectory is clear: the same skim data already being generated for species inventories and biodiversity surveys can now yield population-level diversity estimates as a byproduct. For a field that has long had to choose between genomic depth and geographic breadth, DipSkmer offers a way to have both, turning shallow sequencing into a quantitative window on the genetic variation that sustains populations.

Subject of Research: Reference-free estimation of population genetic diversity from low-coverage diploid genome skimming data

Article Title: DipSkmer: Reference-free population genomics with diploid genome skims

Article References: Charvel, E., Alves Monteiro, H. J., Mirarab, S., & Bafna, V. (2026). DipSkmer: Reference-free population genomics with diploid genome skims. PLOS Computational Biology, 22(10), e1014823. https://doi.org/10.1371/journal.pcbi.1014823

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014823

Keywords: genome skimming, population genomics, coalescent theory, genetic diversity, alignment-free methods, k-mer analysis, Jaccard index, conservation biology, essential biodiversity variables, sequencing error, diploid genomes, reference-free