Machine Learning and Feature Selection for the Classification of Mental Disorders from Methylation Data

Psychiatric disorder diagnoses are heavily reliant on observable symptoms and clinical traits, the skill level of the physician, and the patient's ability to verbalize experienced events. Therefore, researchers have sought to identify biological markers that accurately differentiate mental disorder subtypes from psychiatrically normal comparison subjects. One such putative biomarker, DNA methylation, has recently become more prevalent in genetic research studies in oncology. This paper proposes to apply this paradigm in a study of the diagnostic accuracy of DNA methylation signatures for classifying schizophrenia, bipolar disorder, and major depressive disorder. Very high classification performance measures were obtained from differentially methylated positions and regions, as well as from selected gene signatures. This work contributes to the path toward the identification of biological signatures for mental disorders.