The reference shapes the signal: chrX genotype summaries and classifier sex disparity between GRCh38 and T2T-CHM13 callsets.

Laboratories that move from GRCh38 to T2T-CHM13 take on a new callset as well as a new assembly. I asked whether the move changes sex-chromosome genotype summaries, and whether it changes how well classifiers trained on those summaries serve males and females. I compared the published 1000 Genomes GRCh38 and T2T-CHM13 callsets for 3,202 people. The two callsets come from different alignment, calling, phasing and filtering workflows, so every difference below belongs to the reference and its pipeline together. In women, the fraction of non-reference chrX genotypes that were heterozygous was 0.605 on GRCh38 and 0.644 on T2T. It fell in African-ancestry women and rose in the other four superpopulations. The rise came from fewer non-reference homozygous genotypes (- 7,757 per woman) and not from more heterozygous ones (- 1,504). Five classifier families were trained on six chrX summaries to predict superpopulation, and the male-female gap in balanced accuracy was compared between callsets. Women were classified better than men under both callsets. Logistic regression, random forest and XGBoost widened that gap on T2T, and the support vector machine and the perceptron narrowed it. After correction across all 20 tests, random forest, XGBoost and the support vector machine remained significant. Separating relatives did not remove the disagreement. Classifiers built from the first 500 common SNPs per chromosome in file order showed no change after correction. The gap measures model performance, not biological sex bias. Resolving the assembly from its callset requires processing the same reads through one pipeline against both references.