Interpreting the molecular and cellular landscape of PCOS through bulk transcriptomics, single-cell transcriptomics and machine learning

<h4>Background</h4>Polycystic ovary syndrome (PCOS) is a prevalent endocrine-metabolic disorder in women, hallmarked by hyperandrogenism, anovulation, and polycystic ovarian morphology. This study integrates multi-omics and machine-learning analyses to elucidate the molecular mechanisms and cellular constituents underlying PCOS, aiming to uncover potential therapeutic targets and enhance diagnostic precision.<h4>Methods</h4>Bulk and single-cell RNA sequencing identified key granulosa cell subpopulations and gene expression patterns in PCOS; subsequently, machine-learning algorithms were applied to construct a diagnostic model and to screen for key gene signatures. Consequently, the identified signatures were validated at both mRNA and protein levels in independent clinical samples using qPCR and western blotting.<h4>Results</h4>Compared with controls, PCOS patients exhibited a markedly increased proportion of the GC9 granulosa cell subset, which displayed an active proliferative phenotype. Up-regulated genes in PCOS were closely associated with immune function, responsiveness to stimuli, and diverse cellular biological processes. Machine-learning analysis further pinpointed a three-gene signature-comprising HLA-DRA, SRM, and CTSL-and yielded a diagnostic model with superior accuracy and specificity. Moreover, validation in clinical samples confirmed significant up-regulation of HLA-DRA, SRM, and CTSL at both mRNA and protein levels in follicular cells of PCOS patients.<h4>Conclusions</h4>Our findings delineate a previously unrecognized cellular landscape and gene signature associated with PCOS, thereby proposing novel diagnostic and therapeutic targets.