Endocrine Abstracts

The pathophysiology of Polycystic Ovary Syndrome (PCOS) is multifactorial, therefore discovering effective treatments is challenging. Bioactive food molecules are a potential avenue for PCOS treatment; however, they often lack robust evidence. Applying machine learning (ML) to a genomic dataset may provide accelerated discovery of molecules and drugs that potentially alleviate symptoms through interactions with PCOS-related genes. 17,600 genes, 2,100 bioactive molecules found in foods, and 1,508 clinically available drugs were collected from open-source datasets. 13 genes associated with PCOS (AMH, AMHR2, AR, FSHR, GNRHR, CYP21A2, LMNA, INSR, DNAH11, BMP15, GDF9, LHB, DLK1) were considered as targets to determine the interactivity between PCOS and identified molecules. A higher ranking implied greater interaction between pathways associated with target genes and bioactive molecules (or drugs). Fisher’s exact test assessed whether the molecule rankings were better than a random baseline. Findings were validated using prior literature regarding molecules, foods, or drugs identified to significantly improve PCOS symptoms in women. A higher proportion of beneficial targets were in the top 10 compared to the bottom 10 of 2,100 molecules (P = 0.0031). Of the top 10, isoflavones were predicted to interact with AR and reported to possess antiandrogenic properties. Furthermore, levoglutamide and anthocyanidins demonstrated anti-inflammatory properties in animal studies. Newly identified molecules included epicatechin-3-gallate (found in green tea) and 24-methylenecycloartan-3-ol (a triterpenoid found in almonds), recognised to reduce androgens and inflammation. The identified pharmacological agents in the top three drug classes acted on GNRHR, AR, and INSR, which are known to be important in the pathophysiology of PCOS, thus validating the reliability of the algorithm. Our findings show scope in rapid hypothesis generation for nutritional interventions, or clinical drug repurposing, targeted at PCOS. This ML pipeline can be utilised in other polygenic reproductive and/or metabolic disorders to aid in the identification of novel targets.