A Multiclass Support Vector Machine Model for the Identification of Sperm Motility Patterns in Heterogeneous Populations.

Abstract

Sperm motility and the capacity to undergo hyperactivation in the female reproductive tract are requirements for normal fertilization in mammals. Quantitative analyses of motility in response to perturbations that modulate sperm metabolism or key signaling events would be facilitated by a method that quickly and accurately distinguishes different types of motility in large populations of sperm. To that end, we generated a multiclass support vector machine (SVM) model that identifies hyperactivated sperm and four other distinct patterns of sperm motility within a heterogeneous population. Our multiclass model is based on 2,434 sperm tracks from twelve CD1 mice that were captured using a Hamilton-Thorne CEROS instrument for computer-assisted sperm analysis (CASA) and visually classified into one of six distinct motility groups: progressive, intermediate, hyperactivated, slow, weakly motile, and bad (noise). The CASA parameters from these classified tracks were used to develop a series of SVM equations using established algorithms, yielding a model that identifies uncharacterized tracks with an overall accuracy of 86%. The model incorporates these equations into a binary decision tree that sequentially sorts tracks into distinct populations. Five standard CASA measurements are factored into each equation, including average path velocity (VAP), straight-line velocity (VSL), curvilinear velocity (VCL), amplitude of lateral head displacement (ALH) and beat cross frequency (BCF). The first equation identifies all CASA tracks as either vigorous or non-vigorous. Vigorous tracks are then classified as progressive, intermediate or hyperactivated, while non-vigorous tracks are classified as slow, weakly motile, or bad. We tested our multiclass model by comparing motility profiles of sperm incubated for 2 h in HTF medium with and without bicarbonate. After capacitation for 90 min in the presence of bicarbonate, 20-30% of sperm tracks were classified as hyperactivated and ~10% as intermediate. Without bicarbonate, >65% of sperm tracks remained progressive throughout the incubation and <2% were classified as hyperactivated or intermediate, as expected. The model also accurately categorized tracks from mice with defects in sperm motility. For example, sperm from mice that lack GAPDHS, an essential sperm glycolytic isozyme, showed no progressive motility and >96% were classified as slow or weakly motile. We also assessed the motility profiles of C57BL6/J, 129S1/SvlmJ, and PWK/PhJ sperm over a 2 h in vitro capacitation period and found significant differences among these strains, both in the proportions of progressive and slow sperm and in the percentage of sperm that undergo hyperactivation. By enabling detailed, quantitative comparisons of motility throughout capacitation, this model will be a valuable tool for assessing potential genetic, biomolecular, and pharmaceutical effects on sperm motility in large, heterogeneous populations. Supported by the Eunice Kennedy Shriver NICHD/NIH through U01 HD60481 and cooperative agreement U54 HD35041 as part of the Specialized Cooperative Centers Program in Reproduction and Infertility Research. (platform)