Modeling the menstrual cycle stages in the identification of endometriosis biomarkers

Abstract

Endometriosis is a gynecological disorder where the endometrial tissue that lines the inside of the uterus also grows outside it, often causing severe pain and infertility. Approximately 10% of women worldwide have endometriosis, although prevalence statistics are hindered by the lack of satisfactory non-invasive diagnostic methods. Researchers have long sought biomarkers for early, non-surgical diagnosis, yet no potential biomarkers have been approved. One obstacle is the lack of studies accounting for the fluctuation of gene expression throughout the menstrual cycle stages. This study's objective is to make a comparative analysis of gene expression in endometriosis lesions and in endometrium samples and to evaluate the transcriptomic impact of the menstrual cycle stages. The 150 expression samples were obtained from the Gene Expression Omnibus (GSE141549). Samples were separated by the menstrual stage in which they were recollected (proliferative and secretory). To determine the most relevant genes, four feature selection methods were implemented – recursive feature elimination (RFE), gain ratio, one R, and symmetry uncertainty – and their performance was evaluated using accuracy and area under the curve using Random Forest classifier. Preliminary results showed nine genes (MEOX1, MEOX2, C7, LCN6, LOC375295, SCN4B, C18orf34, LTC4S, and EBF3) to be differentially expressed across all groups. Stage-specific genes were also identified, revealing more differentially expressed genes in the proliferative stage than in the secretory. This work is currently confirming these putative markers across other datasets using a consensus machine learning approach which will provide insight on possible biomarkers and their potential in diagnosis and treatment.