Advances in sequencing technologies enable the recovery of large numbers of SNPs from degraded and low-input forensic samples, allowing increasingly fine-scale inference of biogeographical ancestry (BGA). While unsupervised methods such as PCA or STRUCTURE can reveal fine-scale population structure, their outputs do not directly provide the discrete, geographically interpretable groups required for forensic reporting. Furthermore, the continental groupings commonly used for small BGA panels are often too coarse to reflect the finer-scale structure captured by high-density SNP data. Manual definition of reference groups is possible but remains subjective and requires dataset-specific adaptation due to the variable SNP recovery characteristic of degraded forensic samples. Consequently, the resulting groupings may fail to reflect the underlying genetic structure. This motivates automated approaches that derive population groupings directly from genetic data to enable downstream inference.

To address this, we present ReBin, an automated unsupervised clustering pipeline for defining reference population groups from genetic data using the Leiden community detection algorithm, with integrated visualization of inferred groups on a world map. The pipeline was evaluated using the Human Origins dataset from the Allen Ancient DNA Resource (approximately 600k SNPs and 7,000 individuals). To mimic variability in the number of recovered markers, SNPs were randomly subsampled at 12 set sizes (100–500,000 markers), with 30 replicate runs per set size. Clustering stability and cluster granularity were assessed across Leiden resolution parameters, SNP set sizes, and replicate runs.

A single resolution parameter yielded well-resolved clusters across all SNP set sizes, with clustering highly reproducible across replicate runs (adjusted Rand index > 0.86). Small SNP sets recovered broad continental structure (6–7 clusters), whereas increasing marker set sizes revealed progressively finer regional differentiation. Application to forensic SNP panels (VISAGE Basic Tool and ForenSeq Kintelligence Kit) produced geographically coherent reference groups, which were subsequently used for BGA inference with GENOGEOGRAPHER. For the VISAGE Basic Tool, ReBin recovered the expected continental structure, and inference with GENOGEOGRAPHER highlighted known challenges in separating West Asian and European groups.

Overall, ReBin is a robust and data-driven pipeline for defining forensic ancestry reference groups across SNP datasets of varying size, from established BGA panels to datasets with variable SNP recovery, such as those generated from whole-genome sequencing. By deriving population groupings directly from genetic structure, it reduces reliance on manual curation and enables reproducible ancestry inference that adapts to the resolution of the underlying data.

Authors

• Peter Resutik (Zurich Institute of Forensic Medicine, University of Zurich, Switzerland)
• Janine Biner (Signal and Information Processing Laboratory, ETH Zurich, Switzerland)
• Kris van der Gaag (Division of Biological Traces, Netherlands Forensic Institute, Netherlands)
• Natasha Arora (Zurich Institute of Forensic Medicine, University of Zurich, Switzerland)