Robust imputation-based method for eye, hair, and skin colour prediction from low-coverage ancient DNA

TL;DR

This paper introduces a new method to predict eye, hair, and skin color from low-quality ancient DNA using imputation techniques.

Contribution

The novel contribution is a robust imputation-based framework called aHISPlex for accurate phenotype prediction from degraded ancient DNA.

Findings

Phenotype predictions from ancient DNA can be accurate even at low genome coverages (0.1–0.5×).

The aHISPlex framework includes tools and reference datasets for efficient analysis of degraded DNA samples.

Factors like genome coverage and postmortem damage significantly affect imputation efficiency.

Abstract

The prediction of externally visible traits (eye, hair, and skin colours) from DNA can provide valuable information for contemporary and ancient human populations. The validated HIrisPlex-S method is the primary tool in forensics for phenotyping modern samples. The HIrisPlex-S multiplex PCR assay can handle trace DNA from modern samples, but the analysis of degraded, low-coverage ancient DNA (aDNA) presents additional challenges. Genotype imputation has recently proven successful in effectively filling in missing information in aDNA sequences. To assess the feasibility of this approach, we evaluated how key factors, such as genome coverage, minor allele frequency, extent of postmortem damage, and the population origin of the test individual, influence the efficiency of imputing HIrisPlex-S markers and predicting phenotypes. We used high-coverage sequence data from modern individuals and ancient remains for the evaluation. Our results demonstrate that even with genome coverages as low as 0.1–0.5×, the proposed workflow can predict phenotypes from degraded ancient (or forensic) whole genome sequence (WGS) data with good accuracy. To aid the archaeogenetics community, we have developed aHISPlex. This user-friendly, easily deployable imputation-based framework includes the new bioinformatics tools and the pre-made reference data sets required for the whole analysis.

The online version contains supplementary material available at 10.1038/s41598-026-38372-3.

Figures