Machine Learning-Based Approach Highlights the Use of a Genomic Variant Profile for Precision Medicine in Ovarian Failure

Abstract

Ovarian failure (OF) is a common cause of infertility usually diagnosed as idiopathic, with genetic causes accounting for 10–25% of cases. Whole-exome sequencing (WES) may enable identifying contributing genes and variant profiles to stratify the population into subtypes of OF. This study sought to identify a blood-based gene variant profile using accumulation of rare variants to promote precision medicine in fertility preservation programs. A case–control (n = 118, n = 32, respectively) WES study was performed in which only non-synonymous rare variants <5% minor allele frequency (MAF; in the IGSR) and coverage ≥ 100× were considered. A profile of 66 variants of uncertain significance was used for training an unsupervised machine learning model to separate cases from controls (97.2% sensitivity, 99.2% specificity) and stratify the population into two subtypes of OF (A and B) (93.31% sensitivity, 96.67% specificity). Model testing within the IGSR female population predicted 0.5% of women as subtype A and 2.4% as subtype B. This is the first study linking OF to the accumulation of rare variants and generates a new potential taxonomy supporting application of this approach for precision medicine in fertility preservation.