Proteomics-Enabled Deep Learning Machine Algorithms Can Enhance Prediction of Mortality.

Area of Science:

• Cardiovascular Medicine
• Biomarkers
• Machine Learning

Background:

• Individualized risk prediction is crucial for advancing personalized medicine.
• Cardiovascular diseases remain a leading cause of mortality worldwide.
• Accurate risk stratification is essential for effective patient management.

Purpose of the Study:

• To compare the efficacy of proteomics-enabled machine learning (ML) algorithms against classical and clinical risk prediction methods.
• To evaluate the prediction performance for all-cause mortality in patients with cardiovascular risk factors.
• To validate the findings in an independent cohort.

Main Methods:

• Utilized the OLINK-Cardiovascular-II panel to measure 92 proteins in 1,998 individuals (LIFE-Heart Study) and 772 subjects (PLIC cohort).
• Developed protein-based mortality prediction models using eXtreme Gradient Boosting (XGBoost) and a neural network.
• Compared ML models against classical risk scores (Framingham, Systemic Coronary Risk Evaluation) and regression models (logistic, Cox).

Main Results:

• Machine learning models demonstrated significantly higher predictive accuracy (AUCs ranging from 0.83 to 0.94) compared to classical methods (AUCs ranging from 0.55 to 0.67).
• Proteomics-driven XGBoost and neural network models showed superior performance in both internal and external validation cohorts.
• P-values < 0.001 indicate a statistically significant difference between modern ML and classical approaches.

Conclusions:

• Machine learning-driven multiprotein risk models significantly outperform traditional regression models and clinical scores.
• These advanced models provide enhanced prediction of all-cause mortality in individuals at increased cardiovascular risk.
• The findings support the integration of proteomics and ML for improved personalized cardiovascular risk assessment.