Background

Forensic samples collected at crime scenes often exhibit low quality, characterized by low data yield and high genotype errors, which significantly reduces the accuracy of kinship inference. This study aims to systematically compare the performance of different kinship inference methods on low-coverage whole-genome sequencing (WGS) data and to evaluate the improvement achieved through genotype imputation.
Methods: Sixty-nine samples from a large Chinese Han family were sequenced at 30× and were downsampled to 0.1×, 0.25×, and 0.5× to simulate low-coverage sequencing data. Genotype imputation was performed using GLIMPSE2. Kinship inference was conducted on data before and after imputation using two methods of moment (MoM)-based approaches, KING and READv2, as well as three IBD-based methods, IBIS, TRUFFLE, and clusIBD. The inference accuracies were compared for relationships ranging from first to seventh degree.

Results

Genotype concordance rates decreased with lower sequencing depth when compared to the 30× high-depth data: 72.1% at 0.1×, 75.3% at 0.25×, and 77.5% at 0.5×. After imputation, these rates increased to approximately 88.1%, 94.7%, and 97.1%, respectively. The predominant error pattern was allele dropout. For kinship inference prior to imputation, READv2 yielded the best results across all three depth groups, followed by KING. In contrast, TRUFFLE failed to detect any IBD segments, and the majority of IBD segments detected by IBIS and clusIBD at 0.25× and 0.1× were false positives. After imputation, the accuracy improved substantially. Notably, clusIBD performed the best among the five methods ( 59.97% at 0.25× and 74.04% at 0.5×), while IBIS performed the worst (5.88% at 0.5× and 7.29% at 0.25×). At the lowest-coverage group (0.1×), MoM-based methods still outperformed IBD-based methods. 

Conclusion

This study demonstrates that for low-coverage WGS data, imputation substantially improves genotype accuracy; however, the accuracy of kinship inference varies greatly depending on the methods used. Among the five methods, clusIBD performs best at coverage ≥0.25× after imputation, whereas READv2 achieves high accuracy even at very low depths (0.1×). These results provide a reference for method selection in forensic kinship analysis: READv2 is suitable for ultra-low coverage, whereas clusIBD is preferred when coverage reaches ≥0.25× with imputation.

Authors

• Tingjun Li (Faculty of Forensic Medicine, Zhongshan School of Medicine, Sun Yat-sen University, China)
• Shiyun Meng (Faculty of Forensic Medicine, Zhongshan School of Medicine, Sun Yat-sen University, China)
• Ran Li (Faculty of Forensic Medicine, Zhongshan School of Medicine, Sun Yat-sen University, China)
• Hongyu Sun (Faculty of Forensic Medicine, Zhongshan School of Medicine, Sun Yat-sen University, China)