A machine learning-based clinical tool for diagnosing myopathy using multi-cohort microarray expression profiles.

Area of Science:

• Biomedical Informatics
• Molecular Biology
• Genetics

Background:

• Myopathies are diverse skeletal muscle disorders frequently challenging to diagnose in primary care.
• Molecular expression profiles offer potential for accurate disease diagnosis.
• A novel machine learning tool is proposed for predicting muscle disease subtypes.

Purpose of the Study:

• To develop and validate a machine learning tool for classifying myopathy subtypes.
• To utilize multi-cohort microarray gene expression data for prediction.
• To address diagnostic challenges in primary care for myopathies.

Main Methods:

• Analysis of gene expression data from 1260 patients across 42 cohorts.
• Categorization of cohorts into five muscle disease subtypes: immobility, inflammatory, intensive care unit acquired weakness (ICUAW), congenital, and chronic systemic disease.
• Training Support Vector Machine (SVM) models on selected gene signatures using analysis of variance (ANOVA) and validating with area under the receiver operator curve (AUC).

Main Results:

• The machine learning tool demonstrated varying performance across subtypes, with chronic systemic disease best predicted (AUC 0.872).
• Intensive care unit acquired weakness (ICUAW) and immobility were the least discriminated classes (AUCs 0.777 and 0.789, respectively).
• Gene set enrichment analysis identified relevant biological processes, including neural precursor cell proliferation for ICUAW and aerobic respiration for congenital myopathies.

Conclusions:

• The developed tool is a well-performing molecular classification method for muscle diseases.
• Selected gene markers show promise for clinical application in myopathy diagnosis.
• This tool addresses a significant gap in myopathy diagnosis, offering potential clinical utility.