Related Experiment Video
Updated: May 14, 2026
04:41
Digital PCR for Quantifying Circulating MicroRNAs in Acute Myocardial Infarction and Cardiovascular Disease
Published on: July 3, 2018
Development and Validation of a SHAP-Interpretable Machine Learning Model for Stroke Risk Prediction Using
Qiu Yao1, Bangqing Sun1, Jingyu Shen1
1Nanxiang Branch of Ruijin Hospital, Shanghai, 201802, China.
Journal of Molecular Neuroscience : MN
|May 12, 2026
Summary
This study identifies serum microRNAs (miRNAs) as accurate biomarkers for predicting stroke risk using machine learning. The developed model shows potential for non-invasive stroke detection and risk stratification.
Area of Science:
- Biomarkers and diagnostics
- Genomics and transcriptomics
- Machine learning in healthcare
Background:
- Stroke is a leading cause of death and disability globally.
- Circulating microRNAs (miRNAs) show promise as non-invasive biomarkers for cardiovascular diseases.
- Challenges exist in using miRNAs for stroke prediction due to feature selection and model interpretability.
Purpose of the Study:
- To develop an interpretable machine learning framework for predicting stroke incidence using serum miRNA signatures.
- To identify robust and clinically generalizable miRNA biomarkers for stroke.
- To enhance the interpretability of predictive models for stroke risk.
Main Methods:
- Analysis of serum miRNA expression profiles from 1,785 human samples (GEO dataset GSE117064).
- Differential expression analysis and LASSO logistic regression for feature selection.
- Evaluation of five machine learning classifiers (SVM, Random Forest, XGBoost, etc.) with 10-fold cross-validation.
- Interpretation of the best-performing SVM model using SHAP analysis.
- Validation using quantitative real-time PCR (qPCR) and Next-Generation Sequencing (NGS) on independent cohorts.
Main Results:
- 604 differentially expressed miRNAs identified; 66 selected as predictive features by LASSO regression.
- Support Vector Machine (SVM) model achieved high accuracy (0.9983) and AUC (1.0000).
- SHAP analysis identified key miRNAs (e.g., hsa-miR-3648, hsa-miR-1290) influencing stroke prediction.
- External validation using NGS showed 80% accuracy, supporting cross-platform generalizability.
Conclusions:
- An interpretable machine learning framework integrating miRNA expression, feature selection, and SHAP analysis effectively identifies stroke predictive signatures.
- External validation confirms the robustness and cross-platform potential of identified miRNA biomarkers.
- This approach supports the development of non-invasive diagnostic tools for stroke risk prediction.