Benchmarking KinSNP®: A study on genetic relationship prediction for forensic applications.

Accurate kinship inference is a cornerstone of forensic DNA analysis, particularly in cases involving unidentified human remains in the absence of direct reference samples and when antemortem samples lack reliable chain of custody. This study evaluates the performance of KinSNP®, a forensic software tool developed by Othram, for predicting genetic relationships using whole genome single nucleotide polymorphism (SNP) data. Using a well-characterized family cohort of 12 individuals with known relationships extending to the 6th degree, we benchmarked KinSNP® against GEDmatch PRO™ and the Shared cM Project 4.0 tool. The use of real pedigree data is complementary to previous work assessing the software against synthetic data [1] Relationship predictions were assessed across 66 pairwise comparisons using shared centimorgan (cM) thresholds of 7 and 12 cM. KinSNP® correctly identified 82 % of known relationships within the highest predicted relationship category, and a further 12 % within the second highest predicted relationship category, demonstrating comparable accuracy to existing tools. However, predictive confidence decreased for distant relationships (≥5th degree), and spurious matches were observed among unrelated individuals, particularly at lower cM thresholds. However, increasing the minimum threshold to 12 cM reduced false positives without compromising sensitivity for close relationships. KinSNP® is part of secure desktop application, SNPSuite, developed for advanced SNP-based DNA analysis, where the tools available also accurately predicted biological sex and biogeographical ancestry, aligning with self-reported data. Whilst representing a relatively small sample cohort, these findings provide further evidence to support the utility of KinSNP® as a secure, offline solution for SNP-based kinship analysis in forensic contexts, including disaster victim identification and long-term missing person investigations, where either an STR profile cannot be obtained, or there are no direct or close family STR reference samples with which to compare.