SNP-based prediction of schizophrenia using machine learning

Zamart Ramazanova, Bakhyt T. Matkarimov, Sheida Nabavi, Byron Lawrence Crape, Amin Zollanvari

Bioinformatics Advances, 2025

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.1093/bioadv/vbaf219
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Motivation: Schizophrenia is a complex psychiatric disorder characterized by the deterioration of intellectual processes and emotional responses, affecting ∼0.32% of the global population. The study of single nucleotide polymorphisms (SNPs) associated with schizophrenia is crucial to identifying pathogenic genetic variants and understanding the genetic architecture of this complex disorder. This study aims to demonstrate the feasibility of predicting schizophrenia using an individual's SNP profile. Results: We used genome-wide association (GWA) schizophrenia data from a case-control study across European-American (EA) and African-American (AA) populations, consisting of 4693 participants (46.1% diagnosed with schizophrenia). Machine learning techniques were employed to construct SNP-based predictive models specific to ethnicity-gender groups (EA-F, EA-M, AA-F, and AA-M) where "F" and "M" identify female and male populations, respectively. Feature selection and association analysis were utilized to rank and detect significantly associated SNPs. Model selection was based on stratified five-fold cross-validation. Our EA-F-, EA-M-, AA-F-, and AA-M-specific models achieved classification accuracies (AUC, sensitivity, specificity) of 75.1% (69.2%, 95.4%, 34.2%), 65.4% (74.3%, 73.6%, 58.6%), 68.6% (69.5%, 85.6%, 41.1%), and 73.9% (74.0%, 39.7%, 93.3%), respectively, in independent test sets from the same ethnicity-gender population. The high sensitivity (>70%) of AA-F, EA-F, and EA-M models can make them auxiliary clinical tools to assess the risk of developing schizophrenia disorder in AA-F, EA-F, and EA-M populations. Availability and implementation: The R and Python codes, along with the dataset used in this study are available at https://github.com/Zam-gif/Prediction-Models-of-Schizophrenia/tree/main.