Integrated machine learning analysis of proteomic and transcriptomic data identifies healing associated targets in diabetic wound repair

This study identified biomarkers that could be leveraged to classify the state of healing in diabetic wounds. Firstly, by collecting wound samples from diabetic mice at different time points and generating their protein profiles using standard techniques, we set to interrogate whether a small number of biomarkers could serve as sensors to monitor the healing stage. Least absolute shrinkage and selection operator (LASSO) was applied. Large-scale analysis of wound tissue proteins integrated with the respective wound sizes allowed to establish a correlation between the observed protein profile and wound closure. We further evaluated human subjects' systemic serum proteomics for biomarkers. An additional wound healing model in diabetic mice was employed for microRNA quantitation at the same time points and similarly analyzed. Our analysis highlighted markers MMP-2, HGF, miR-1b and miR-107-3p in mice and Fractalkine and FGF-2 in humans that could correctly identify the extent of healing. By using proteomics from mice and human patients and complementary microRNA mouse data with computer regression models we can better predict molecular and protein deficits associated with impaired diabetic wound repair.