Real-Time Precision Psychiatry for Schizophrenia: A Microfluidic Genotyping and Super-Learner Platform for Treatment Response Prediction.
Antipsychotic treatment response in schizophrenia shows substantial interindividual variability. While pharmacogenomics (PGx) holds potential for personalized prescribing, current models fall short in integrating multimodal factors and meeting real-time clinical decision-making needs.
The precision medicine platform integrating multi-dimensional factors can provide guidance for selecting antipsychotic medications for patients with schizophrenia.
This study employed a retrospective cohort design for model development (2019-2021, N = 735) and external validation (2021-2022, N = 90). We integrated PGx profiles, clinical characteristics, and medication data, selected features via random forest recursive feature elimination (RF-RFE), and built a Super Learner ensemble model (incorporating 5 machine learning algorithms) to predict Positive and Negative Syndrome Scale (PANSS) reduction. For 13 prioritized single-nucleotide polymorphisms (SNPs), we developed a kompetitive allele-specific PCR-based microfluidic chip and validated its accuracy against Sanger sequencing in 24 clinical samples. A clinical decision support (CDS) tool integrating genotyping and predictive analytics was deployed.
The Super Learner model achieved a cross-validated RMSE of 7.08 (R2 = 0.89) and an external validation RMSE of 9.02 (R2 = 0.76). The microfluidic chip showed 100% concordance with Sanger sequencing across all 13 SNPs. The integrated CDS system demonstrated sample-to-report feasibility within 3 h.
Microfluidic genotyping and Super Learner platform enables rapid, real-time prediction of antipsychotic treatment response in schizophrenia, providing a scalable tool for early personalized treatment selection during hospitalization.
The precision medicine platform integrating multi-dimensional factors can provide guidance for selecting antipsychotic medications for patients with schizophrenia.
This study employed a retrospective cohort design for model development (2019-2021, N = 735) and external validation (2021-2022, N = 90). We integrated PGx profiles, clinical characteristics, and medication data, selected features via random forest recursive feature elimination (RF-RFE), and built a Super Learner ensemble model (incorporating 5 machine learning algorithms) to predict Positive and Negative Syndrome Scale (PANSS) reduction. For 13 prioritized single-nucleotide polymorphisms (SNPs), we developed a kompetitive allele-specific PCR-based microfluidic chip and validated its accuracy against Sanger sequencing in 24 clinical samples. A clinical decision support (CDS) tool integrating genotyping and predictive analytics was deployed.
The Super Learner model achieved a cross-validated RMSE of 7.08 (R2 = 0.89) and an external validation RMSE of 9.02 (R2 = 0.76). The microfluidic chip showed 100% concordance with Sanger sequencing across all 13 SNPs. The integrated CDS system demonstrated sample-to-report feasibility within 3 h.
Microfluidic genotyping and Super Learner platform enables rapid, real-time prediction of antipsychotic treatment response in schizophrenia, providing a scalable tool for early personalized treatment selection during hospitalization.
Authors
Li Li, Tang Tang, Yu Yu, Wang Wang, Feng Feng, Qu Qu, Hu Hu, Zhao Zhao, Cao Cao, Xue Xue
View on Pubmed