An integrated network topology and deep learning model for prediction of Alzheimer disease candidate genes

Alzheimer’s disease (AD) is a neurological illness that causes short-term memory loss. There are currently no viable therapeutic therapies for this condition that can cure it. The source of Alzheimer’s disease is unknown. However, genetic factors are thought to have a role in the illness’s development, with about 70% of the disease’s risk attributed to the vast number of genes associated. Despite discovering several potential AD susceptibility genes through genetic association studies, there is a more significant challenge to identify unidentified AD-associated genes and drug targets to gain a good insight into the disease-causing mechanisms of Alzheimer’s disease and develop effective AD therapeutics. The proposed DC-GC (Degree Centrality- Graph Colouring) model brings an accuracy of 96% for (Artificial Neural Network) ANN model, 87.3% for KNN (K-Nearest Neighbourhood classifier) classifier, 86% for SVM (Support Vector Machine) classifier, 85.3% than Decision Tree. It is visible; the network topology model performs well for ANN classifier than other existing models. Similarly, the model also brings a sensitivity measure of 97% for the ANN model, 84% for KNN (K-Nearest Neighbourhood classifier), 84.2% for SVM (Support Vector Machine) classifier and 84% for the Decision tree classifier. In this research work, a novel network topology measure DC-GC (Degree Centrality- Graph Colouring) and intelligent-based machine learning models are used for identifying candidate genes from protein–protein interaction and sequence features of genes. The integrated method helps to identify the target gene for Alzheimer’s disease by evaluating the connectivity between the genes and the physicochemical properties of the genes. The approach helps to rank the genes according to the property that adjacency genes should not share the same colour. The DC-GC (Degree Centrality-Graph Colouring)-based network topology measure provides remarkable improvement over existing centrality measures. The integration of network topology measure with the SVM (Support Vector Machine) model gave promising results of 96% accuracy, 97% sensitivity, 98% specificity, 96% PPV (Positive Predictive Value), 95% NPV (Negative Predictive Value) and 97% F-score.