Role of different omics data in the diagnosis of schizophrenia disorder: A machine learning study

Abstract

Schizophrenia is a serious mental disorder that affects millions of people worldwide. This disorder slowly disintegrates thinking ability and changes behaviours of patients. These patients will show some psychotic symptoms such as hallucinations, delusions, thought disorder and movement disorder. These symptoms are in common with some other psychiatric disorders such as bipolar disorder, major depressive disorder and mood spectrum disorder. As patients would require immediate treatment, an on-time diagnosis is critical. This study explores the use of omics data in diagnosis of schizophrenia. Transcriptome, miRNA and epigenome data are used in diagnosis of patients with schizophrenia with the aid of machine learning algorithms. As the data is in high dimension, mutual information and feature importance are independently used for selecting relevant features for the study. Selected sets of features (biomarkers) are individually used with different machine learning algorithms and their performances are compared to select the best-performing model. This study shows that the top 140 miRNA features selected using mutual information along with support vector machines give the highest accuracy (0.86 ± 0.07) in the diagnosis of schizophrenia. All reported accuracies are validated using 5-fold cross validation. They are further validated using leave one out cross validation and the accuracies are reported in the supplementary material.

Keywords: DNA methylation; Feature selection; Machine learning; Schizophrenia diagnosis; Transcriptome; miRNA.