Advances in Redox Research (Jul 2022)
Identification of severity and passive measurement of oxidative stress biomarkers for β–thalassemia patients: K-means, random forest, XGBoost, decision tree, neural network based novel framework
Abstract
Iron-overload induced oxidative stress in β–thalassemia patients is one of the major challenges in the associated treatment protocol. The severity and accurate prognosis of the disease is largely dependent on several clinical parameters, and different hormones also provide passive indication in this context. The present study focused on clinical implication of iron overload, tropic hormones, and oxidative stress biomarkers on severity of β–thalassemia patients as well as understating their interactive effects. Extensive blood serum samples were collected from group of case and control and statistical analysis was performed on the analyzed study parameters from the serum samples. The oxidative stress biomarkers, Malondialdehyde (MDA) and protein carbonyl level showed significant positive correlation with ferritin levels in case. A novel framework was developed to categorize the severity of the disease through K-means clustering and several classification algorithms, such as XGBoost, random forest, and decision tree. Furthermore, a neural network model was used for predicting the oxidative stress biomarker, MDA and protein carbonyl from measured value of ferritin and trophic hormones. The results of clustering depicted that ferritin and the oxidative stress biomarkers were conclusive parameters in determining the severity of the disease. Among the classifiers, XGBoost showed the highest accuracy after k-cross validation (100%). The neural network model exhibited high accuracy in predicting MDA and protein carbonyl. The proposed technique can be chosen as a real life decision tool for medical professionals in the diagnosis and treatment of β–thalassemia. Furthermore, the approach of passive determination of some critical blood parameters may be attributed from the developed prediction model, which can also be instrumental in the similar area of medical research.
