Biochemistry and Biophysics Reports (Dec 2024)
CCN5/WISP2 serum levels in patients with coronary artery disease and type 2 diabetes and its correlation with inflammation and insulin resistance; a machine learning approach
Abstract
Introduction: Studies have shown various effects of CCN5/WISP2 on metabolic pathways, yet no prior investigation has established a link between its serum levels and CAD and/or T2DM. Therefore, this study seeks to explore the relation between CCN5 and the risk factor of CAD and/or diabetes, in comparison to individuals with good health, marking a pioneering endeavor in this field. Methods: This case-control study investigates serum levels of CCN5, TNF-α, IL-6, adiponectin, and fasting insulin in a population of 160 individuals recruited into four equal groups (T2DM, CAD, CAD-T2DM, and healthy controls). Statistical tests comprise Chi-square tests, ANOVA, Spearman correlation, and logistic regression. ROC curves were used to represent the diagnostic potential of CCN5. Disease states are predicted by machine learning algorithms: Decision Tree, Gradient Boosted Trees, Random Forest, Naïve Bayes, and KNN. These models' performance was evaluated by various metrics, all of which were ensured to be robust by applying 10-fold cross-validation. Analyses were done in SPSS and GraphPad Prism and RapidMiner software. Results: The CAD, T2DM, and CAD-T2DM groups had significantly higher CCN5 concentrations compared to the healthy control group (CAD: 336.87 ± 107.36 ng/mL, T2DM: 367.46 ± 102.15 ng/mL, CAD-T2DM: 404.68 ± 108.15 ng/mL, control: 205.62 ± 63.34 ng/mL; P < 0.001). A positive and significant correlation was observed between CCN5 and cytokines (IL-6 and TNF-α) in all patient groups (P < 0.05). Multinomial logistic regression analysis indicated a significant association between CCN5 and T2DM-CAD, T2DM, and CAD conditions (P < 0.001) even after adjusting for gender, BMI, and age (P < 0.001). Regarding the machine learning models, the Naïve Bayes model showed the best performance for classifying cases of T2DM, achieving an AUC value of 0.938±0.066. For predicting CAD, the Random Forest classifier achieved the highest AUC value of 0.994±0.020. In the case of CAD-T2DM prediction, the Naïve Bayes model demonstrated the highest AUC of 0.981±0.059, along with an Accuracy of 97.50 % ± 7.91 % and an F-measure of 96.67 % ± 10.54 %. Conclusion: Our study has revealed, for the first time, a positive connection between CCN5 serum levels and the risk of developing T2DM and CAD. Nonetheless, more research is needed to ascertain whether CCN5 can serve as a predictive marker.
