Abstract
BACKGROUND: Host sex is a critical determinant of microbial community structure, influenced by hormonal profiles, physiology, and sex-stratified behaviors. Despite its importance, sex metadata is frequently missing or mislabeled in microbiome studies. Existing genomic sex-calling tools often fail in low-host-biomass samples (e.g., stool) because they require high read depths to achieve reliability. RESULTS: Here, we present SCiMS (Sex Calling in Metagenomic Sequences), a bioinformatic tool that leverages host-derived DNA within metagenomic datasets to accurately predict host sex, even at low host coverage. SCiMS uses sex-chromosome read density ratios within a Bayesian classifier to provide high-accuracy sex calls. In simulations, SCiMS achieves >85% accuracy with as few as 450 host reads. When applied to 1,339 samples from the Human Microbiome Project, SCiMS outperforms existing tools, showing higher accuracy and more balanced precision-recall tradeoffs across body sites. SCiMS also generalizes effectively to non-human hosts, achieving 100% accuracy in a murine dataset and outperforming alternatives in a chicken dataset with a ZW sex determination system. CONCLUSIONS: SCiMS provides an accurate, scalable, and cross-species generalizable solution for host sex classification in metagenomic datasets, even when host DNA is minimal. By enabling the recovery of missing sex metadata, it serves as a quality-control tool for ensuring the integrity of analyses in microbiome research. SCiMS is freely available at http://github.com/davenport-lab/SCiMS.