Machine Learning Classifiers for Endometriosis Using Transcriptomics and Methylomics Data.

Area of Science:

• Genomics
• Bioinformatics
• Gynecology

Background:

• Endometriosis is a prevalent yet poorly understood gynecological disorder affecting millions globally.
• Significant diagnostic latency (4-11 years) exists due to lack of definitive symptoms and minimally invasive diagnostic methods.
• Machine learning offers potential for discovering biological patterns in large-scale omics data.

Purpose of the Study:

• To evaluate machine learning methods for classifying endometriosis using transcriptomics and methylomics data.
• To assess the impact of normalization techniques and feature selection on classification performance.
• To identify potential diagnostic biomarkers for endometriosis.

Main Methods:

• Analysis of 38 RNA-seq and 80 DNA methylation (MBD-seq) datasets.
• Application of supervised machine learning algorithms (decision tree, PLSDA, SVM, random forest).
• Comparison of normalization techniques (TMM, quantile, voom) and generalized linear model (GLM) for feature reduction.

Main Results:

• Machine learning models successfully classified endometriosis from control samples.
• Specific normalization strategies (TMM for RNA-seq, quantile/voom for MBD-seq) and GLM improved performance.
• Identified candidate biomarkers including NOTCH3, SNAPC2, TRPM6, and RASSF2.

Conclusions:

• An optimized machine learning pipeline utilizing specific normalization and GLM can enhance endometriosis classification.
• The identified genes represent potential biomarkers for improved endometriosis diagnosis.
• Further validation of these biomarkers is warranted for clinical application.