Machine learning identifies proteomic risk factors across 23 diseases.

Area of Science:

• Biomedical Science
• Proteomics
• Machine Learning in Medicine

Background:

• Early disease detection is vital for effective medical intervention.
• Plasma proteome analysis offers potential for understanding disease and improving patient outcomes.
• Current diagnostic methods can be invasive and lack speed.

Purpose of the Study:

• To develop and validate a novel classifier for multi-disease detection using plasma proteomic data.
• To assess the performance of a domain-expertise-guided hierarchical classifier against traditional machine learning algorithms.
• To explore the utility of plasma proteomics for broad disease screening.

Main Methods:

• Collected plasma proteomic data from over 3000 patients across 23 diseases, analyzing 1462 proteins.
• Developed a two-stage hierarchical classifier integrating histological knowledge.
• Applied the classifier to multi-disease classification and compared its performance with standard machine learning approaches.

Main Results:

• The developed hierarchical classifier demonstrated superior prediction performance compared to traditional machine learning methods.
• The classifier exhibited improved feature selection and a better balance in classification outcomes.
• Empirical guidance from domain expertise significantly enhanced the machine learning model's effectiveness.

Conclusions:

• Integrating domain expertise into machine learning models improves disease detection accuracy.
• Plasma proteomics is a promising tool for multi-disease screening and early diagnosis.
• The developed classifier shows potential for minimally invasive and rapid medical diagnostics.