Building Predictive Models for Schizophrenia Diagnosis with Peripheral Inflammatory Biomarkers

Area of Science:

• Psychiatry
• Computational Neuroscience
• Biomarker Discovery

Background:

• Precision psychiatry leverages machine learning (ML) and artificial intelligence (AI) for analyzing complex, multi-domain data.
• Schizophrenia research commonly utilizes neuroimaging, voice, language, and mobile phone data.
• Peripheral biological markers offer an additional valuable data source for predictive modeling in schizophrenia.

Purpose of the Study:

• To develop and evaluate five distinct predictive models for binary classification of schizophrenia patients versus healthy individuals.
• To assess the utility of serum concentrations of cytokines, chemokines, growth factors, and age, among 38 parameters, for schizophrenia prediction.
• To investigate the diagnostic potential of peripheral biomarkers using machine learning approaches.

Main Methods:

• Developed five predictive models: logistic regression, deep neural networks (DNN), decision trees, support vector machine (SVM), and k-nearest neighbors (KNN).
• Utilized a dataset comprising 38 parameters, including serum biomarker concentrations and age, from 217 schizophrenia patients and 90 healthy controls.
• Evaluated model performance based on sensitivity and specificity for binary classification.

Main Results:

• The deep neural network (five-layer) model demonstrated slightly superior performance with a sensitivity of 0.87 ± 0.04 and specificity of 0.52 ± 0.06.
• Combining all 38 variables into a single classifier yielded a cumulative effect surpassing individual variable effectiveness.
• The findings underscore the necessity of integrating multiple biomarkers for accurate schizophrenia diagnosis.

Conclusions:

• Peripheral biomarker data, when analyzed with machine learning methods, shows significant promise for schizophrenia diagnosis.
• Deep neural networks represent a powerful tool for analyzing complex biomarker data in psychiatric research.
• The study advocates for a multi-biomarker approach, enhanced by ML, for improved diagnostic accuracy in schizophrenia.