The machine learning algorithm for the diagnosis of schizophrenia on the basis of gene expression in peripheral blood

Abstract

BackgroundSchizophrenia (SCZ) is a highly heritable mental disorder with a substantial disease burden. Machine learning (ML) method can be used to identify individuals with SCZ on the basis of blood gene expression data with high accuracy. MethodsThis study aimed to differentiate patients with SCZ from healthy individuals by using the messenger RNA expression level in peripheral blood of 48 patients with SCZ and 50 controls via ML algorithms, namely, artificial neural networks, extreme gradient boosting, support vector machine (SVM), decision tree, and random forest. The expression of six mRNAs was detected using quantitative real-time polymerase chain reaction (qRT-PCR). ResultsThe relative expression levels of GNAI1 (P < 0.001), PRKCA (P < 0.001), and PRKCB (P = 0.021) increased in the SCZ group, whereas those of FYN (P < 0.001), LYN (P = 0.022), and YWHAZ (P < 0.001) decreased in the SCZ group. We generated models with various combinations of genes based on five ML algorithms. The SVM model with six factors (GNAI1, FYN, PRKCA, YWHAZ, PRKCB, and LYN genes) was the best model for distinguishing patients with SCZ from healthy individuals (AUC = 0.993, sensitivity = 1.000, specificity = 0.895, and Youden index = 0.895). ConclusionsThis study suggested that the combination of genes using the ML method is better than the use of a single gene to discriminate patients with SCZ from healthy individuals. The combination of GNAI1, FYN, PRKCA, YWHAZ, PRKCB, and LYN under the SVM model can be used as a diagnostic biomarker for SCZ.