Continuous Motility Fingerprints from Gaussian Mixture Models Reveal Hidden Sperm Heterogeneity

Computer-assisted sperm analysis (CASA) collapses sperm motility into coarse categorical bins, masking the heterogeneity that influences fertilization potential. Analyzing 33,500 trajectories from 340 individuals, we introduce a probabilistic fingerprinting framework that treats Gaussian mixture model (GMM) posteriors as soft supervision to construct continuous motility axes that capture forward progression, erratic movement, and uncertainty. This semi-supervised, posterior-guided approach condenses high-dimensional CASA features into interpretable, uncertainty-aware descriptors that capture both canonical and transitionary sperm behaviors. GMM consistently identified five reproducible subtypes, including an erratic group characterized by vigorous, asymmetric motion associated with fertilization readiness. Patient-level fingerprints revealed robust associations with semen quality markers (debris, round cells, concentration), demonstrating potential utility for fertility evaluation and clinical decision-making. Together, our results provide the first large-scale evidence that probabilistic fingerprints capture biologically meaningful and clinically anchored sperm heterogeneity, establishing a generalizable framework for outcome-linked studies in reproductive medicine.