From fathers to forebears: Benchmarking Y-chromosomal short tandem repeat haplogroup predictors for forensic paternal lineage inference.

The male-specific Y chromosome plays a pivotal role in population and forensic genetics. Its relatively conserved paternal inheritance and lack of recombination over 95% of its length provide valuable insights into paternal lineage ancestry and DNA kinship investigations. While slowly mutating Y-chromosomal single nucleotide polymorphisms (Y-SNPs) are traditionally used to assign evolutionary Y-(sub)haplogroups, the process is labour-intensive, time-consuming and costly. An efficient alternative approach involves the use of the more rapidly mutating familial Y-chromosomal short tandem repeats (Y-STRs). To this end, several Y-STR haplogroup predictors have been developed. Nevertheless, there remains some uncertainty about their prediction accuracy and overall suitability for forensic applications. In the current study, the performance of three Y-STR haplogroup predictors was evaluated, specifically Whit Athey's, the NevGen, and the PredYMaLe Haplogroup Predictor. The validation was based on Y-STR and Y-SNP data from 2193 males in our CSY-database, which is representative for Flemish and Dutch populations. Y-STR haplotypes were entered into the Y-STR haplogroup predictors and the predictions were then compared to the Y-SNP-derived haplogroups, offering a means to assess prediction accuracy. Additionally, the influence of input Y-STR marker-panel variability on prediction accuracy was evaluated using Y-STRs from the commercially available PowerPlex® Y23 System (23 Y-STR loci; Promega Corporation) and Yfiler® Plus Kit (27 Y-STR loci; Thermo Fisher Scientific), and those from our in-house YForGen Kit (38-46 Y-STR loci; UZ/KU Leuven). Whit Athey's programs - especially its Main 111-Marker Program - performed well for assigning the most common European Y-(sub)haplogroups at broader-level depths into the phylogenetic tree (less detailed Y-subhaplogroup level) but showed reduced predictive performance at finer-level depths (more detailed Y-subhaplogroup level) and for poorly represented lineages. By contrast, NevGen with input Y-STRs from the YForGen Kit, and PredYMaLe were favoured at finer-level depths. However, prediction accuracy should not be the sole criterion for selecting an appropriate predictor, as its forensic usability and utility are inherently context dependent. Therefore, Whit Athey's programs and NevGen are best suited for small-scale forensic casework, whereas PredYMaLe is more appropriate for population-scale investigations when representative training data are available. Where possible, Y-SNP-based confirmation remains recommended.