Proteomic machine-learning signatures associated with carotid plaque: A population-based hypothesis-generating study

Carotid plaque (CP) is a surrogate marker of cardiovascular disease, and high-risk morphology significantly increases the risk of stroke. The aim of this study was to develop advanced machine learning (ML) models to predict the presence of CP in the general population. The study involved 693 participants from the Polish Longitudinal University Study - Bialystok PLUS, a representative sample of the local population. The analysis used 460 OLINK protein biomarkers. The data were analysed using the BORUTA algorithm based on the random forest classifier method. We identified 42 significant biomarkers associated with CP, the most important of which were GDF-15, CDCP1, CHIT1, FLT3LG, CDH2 and CRTAC1. Interestingly, the top 6 biomarkers had a highly speculative or unresolved association with atherosclerosis. Next, we developed an ML-based CP fingerprint comprising 25 biomarkers, which showed good internal cross-validated discrimination in subjects aged 30-70 years. We also generated a network of interactions for the 42 biomarkers using STRING database modules, and performed a functional analysis to assess the biological significance of the interactions detected. These results are hypothesis-generating and require external, prospective validation to identify cause-and-effect relationships.