Benchmarking of Machine Learning classifiers on plasma proteomic for COVID-19 severity prediction through...

Area of Science:

• Computational biology and bioinformatics
• Machine learning applications in healthcare
• Infectious disease modeling

Background:

• The COVID-19 pandemic necessitated rapid development of tools for patient severity assessment and triage.
• Existing methods often lack the ability to integrate diverse data types for comprehensive patient evaluation.
• Artificial intelligence (AI) offers potential for analyzing complex biological and clinical data to predict disease outcomes.

Approach:

• An ensemble of machine learning (ML) algorithms was developed to predict COVID-19 patient severity.
• The models utilized plasma proteomics and clinical data as input features.
• An interpretable AI approach using Shapley additive explanation (SHAP) values was employed to identify key predictive factors.

Key Points:

• ML models, particularly Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM), achieved recall scores up to 0.74 and F1-scores up to 0.75.
• Key predictors identified include patient age, B cell dysfunction markers, Toll-like receptor pathway activation, and SCF/c-Kit signaling.
• The models demonstrated the potential to discern critical COVID-19 cases based on a combination of biological and clinical data.

Conclusions:

• The developed AI pipeline shows promise for early COVID-19 patient triage by integrating plasma proteomics and clinical data.
• Interpretable AI analysis revealed specific immuno-biological pathways crucial for predicting disease severity.
• Further validation with larger datasets is recommended to confirm the clinical utility of this AI-driven approach.