Beyond the linear genome: how reference bias threatens preventive medicine and geroscience.

Reference minor alleles in the GRCh38 linear assembly generating false high-impact annotations (SLC37A4, CIMIP2A, GPR33) during automated variant classification; graph-based pangenome references as the proposed way forward.

The GRCh38 linear assembly contains reference minor alleles, a structural vulnerability for automated variant classification whenever a patient genome is compared to that reference. The authors report an observational case series of 20 healthy adults undergoing preventive genome sequencing, processed with an automated pipeline aligned to GRCh38, targeting loci where GRCh38 differs from the population consensus major allele. In all 20 participants (100%), linear alignment systematically misclassified functional major alleles as false-positive high-impact annotations at three distinct loci, in SLC37A4, CIMIP2A and GPR33, these artefacts arising solely because the software mathematically defines the healthy wild-type state as a deviation from the rare reference minor allele; cross-referencing with gnomAD confirmed them as population-dominant benign alleles. The authors further note a theoretical risk of false negatives if such reference alleles mask true pathogenic variants, and argue that downstream population-frequency filters, while managing false positives, amount to an unsustainable patch at population screening scale — hence the appeal of graph-based pangenome references, subject to computational and standardisation challenges.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Reference minor alleles are a long-known phenomenon, already neutralised in serious diagnostic laboratories by population-frequency filters — so the value of this paper is not discovery but the quantified demonstration, across 20 genomes, that an automated pipeline produces the artefact in 100% of healthy subjects. The useful message targets less prescribed diagnostics than direct preventive sequencing and large ageing cohorts, where reporting is automated and expert review rarer. The limitation is clear: three loci described, no comparative evaluation of a pangenome reference, and the false-negative risk remains explicitly theoretical.

Analysis by Dr Thibaut Benquey