Application of a Machine Learning-Based Classification Approach for Developing Host Protein Diagnostic Models for...

Area of Science:

• Biomedical diagnostics
• Computational biology
• Infectious disease research

Background:

• Infectious disease diagnosis traditionally relies on pathogen detection.
• Host-centered approaches offer complementary insights but require complex data interpretation.
• Machine learning (ML) can potentially streamline the analysis of host biomarker data.

Purpose of the Study:

• To develop and present a template for an ML-based classification workflow for host-centered infectious disease diagnosis.
• To build and optimize an ML model for differentiating bacterial, viral, and non-disease states using host protein biomarkers.
• To demonstrate the utility of an automated ML (Auto-ML) approach for identifying diagnostic biomarker signatures.

Main Methods:

• Collected human serum protein data from samples with known disease etiologies (bacterial, viral, normal).
• Employed an automated machine learning (Auto-ML) strategy to train and optimize classification models.
• Validated the performance of the optimized classifier on a blinded set of samples.

Main Results:

• An optimized Auto-ML classification model successfully distinguished between bacterial, viral, and normal human serum samples.
• The model demonstrated robust diagnostic characteristics even with a limited training dataset size.
• Effective performance was observed when the model was applied to an independent, blinded sample set.

Conclusions:

• The presented Auto-ML workflow provides a flexible and adaptable method for developing host biomarker classifiers for infectious diseases.
• This approach can aid researchers in identifying host-based diagnostic signatures for various disease states and biomarker classes.
• Host-centered ML models show promise for complementing traditional pathogen-directed diagnostic methods.