Stability-Based Machine Learning Identifies a Minimal Two-Protein Serum Signature for Early Silicosis.

Area of Science:

• Pulmonary Medicine
• Biomarker Discovery
• Proteomics

Background:

• Silicosis diagnosis is difficult due to low sensitivity of radiological methods in early stages.
• Interobserver variability in radiological assessment further complicates early detection.
• There is a critical need for noninvasive, objective biomarkers for timely silicosis diagnosis and intervention.

Purpose of the Study:

• To identify novel, noninvasive serum biomarkers for early-stage silicosis.
• To develop and validate a machine learning-based diagnostic signature for silicosis.

Main Methods:

• A multistage study design with discovery and validation cohorts.
• Olink targeted proteomics for serum protein profiling.
• A stability-based machine learning framework (Lasso, Random Forest, SVM-RFE) for feature selection and a logistic regression model for diagnosis.

Main Results:

• A two-protein signature (IL8 and CCL3) was identified.
• The signature achieved high diagnostic performance (AUC 0.986 discovery, 0.973 validation).
• Decreased serum levels of IL8 and CCL3 were associated with early silicosis.

Conclusions:

• IL8 and CCL3 represent novel, accurate, noninvasive biomarkers for early silicosis detection.
• This protein signature overcomes limitations of current diagnostic methods.
• The findings highlight a paradoxical shift in circulating chemokines in early silicosis.