Robust Computational Model for Diagnosis of Mitogenic Activated Protein Kinase Leading to Neurodegenerative Diseases

Abstract

Computational modeling is used to develop solutions by formulating and modeling real-world problems. This research article presents an innovative approach to using a computational model, as well as an evaluation of software interfaces for usability In this work, a machine learning technique is used to classify different mitogenic activated protein kinases (MAPK), namely extracellular signal-regulated kinase (ERK), c-Jun amino (N)-terminal kinases (JNK), and mitogenic kinase (MK2) proteins. A deficiency of ERK and JNK leads to neurodegenerative diseases, such as Parkinson's disease, Alzheimer's disease (AD), and prion diseases, while the deficiency of MK2 leads to atherosclerosis. In this study, images from a heat map were normalized, scaled, smoothed, and sharpened. Different feature extraction methods have been used for various attributes, while principal component analysis was used as a feature selection technique. These features were extracted with machine learning algorithms to produce promising results for clinical applications. The results show that ANN achieves 97.09%, 96.82%, and 96.01% accuracy for JNK, ERK, and MK2 proteins, respectively, whereas CNN achieves 97.60%, 97.36%, and 96.81% accuracy for the same proteins. When CNN is used, the best results are obtained for JNK protein, with a training accuracy of 97.06% and a testing accuracy of 97.6%. The proposed computational model is validated using a convolution neural network (CNN). The effect of the hidden layer on different activation functions has been then observed using ANN and CNN. The proposed model may assist in the detection of various MAPK proteins, yielding promising results for clinical diagnostic applications.