Machine learning in schizophrenia genomics, a case-control study using 5,090 exomes.

Area of Science:

• Genetics
• Computational Biology
• Psychiatry

Background:

• Schizophrenia (SCZ) poses a significant public health challenge.
• Early identification of individuals at high risk for SCZ is crucial for timely intervention.
• Current diagnostic methods may not fully capture genetic predispositions.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML) applied to whole exome sequencing (WES) data for identifying SCZ risk.
• To develop a predictive model for SCZ based on genetic variant patterns.
• To uncover potential genetic factors contributing to SCZ pathophysiology.

Main Methods:

• Applied ML algorithms to WES data from 2,545 SCZ cases and 2,545 controls from dbGaP.
• Annotated single nucleotide variants and small insertions/deletions using ANNOVAR (hg19/GRCh37).
• Utilized eXtreme Gradient Boosting (XGBoost) for supervised learning, training on 70% and testing on 30% of the data.

Main Results:

• The XGBoost model achieved high performance: 85.7% accuracy, 86.6% specificity, 84.9% sensitivity, and an AUC of 0.95.
• Identified a subset of genes significantly associated with SCZ risk.
• Analysis of top predictive genes provided insights into SCZ pathophysiology.

Conclusions:

• ML analysis of WES data is a promising approach for predicting SCZ risk.
• The developed predictor demonstrates significant potential for early detection of SCZ.
• This approach may facilitate research into disease-modifying interventions for SCZ.