Optimal gene prioritization and disease prediction using knowledge based ontology structure

ABSTRACT

Prioritizing candidate genes is essential for genome-based diagnostics of various hereditary disorders. Furthermore, it is a difficult task with particular and noisy information about genes, illnesses, and relationships. Although several computer methods for disease gene prioritization have been developed, their efficiency is limited by manually created traits, network architecture, or pre-established data fusion criteria. Hence, this research proposes a unique gene prioritization and disease prediction model. Initially, the gathered information is pre-processed by a data cleaning model. In the proposed gene prioritization phase, the pre-processed data are tokenized. Then a new knowledge-based ontology structure is constructed with the improved skewness-based semantic similarity function. The ensemble classifier is constructed along Recurrent Neural Network (RNN), optimized fuzzy logic, and also Deep Belief Network (DBN) to forecast the gene disorders in the prediction phase. The retrieved features from the feature extraction phase are used to train RNN;while the extracted knowledge bases are used to train the DBN, then the results are fed into the optimized fuzzy logic. The fuzzy logic is the primary indication;its fuzzification function is fine-tuned employing a methodology to improve illness prediction accuracy. A recommended new hybrid system, named as Cauchy's Mutated Corona Virus Optimization Algorithm (CMCOA), is the upgraded version of the CVOA, a typical coronavirus optimization technique. Finally, to evaluate the efficiency of the projected model, a comparison of the suggested and existent models is performed with respect to various measures. In particular, the proposed model has recorded the highest accuracy as 93 % at 60 % of training, which is 42.5 %, 36.1 %, 33.3 %, 41.1 %, 48.5 %, 48.5 %, 9 %, 8 %, 8 %, 8 %, 8 %, and 14.5 % improved over existing models like GCN, GCN [6], SVM, CNN, Bi-LSTM, LSTM, GRU, fuzzy, EC + GOA, EC + SSO, EC + CMBO, EC + SMA and EC + CCVOA, respectively. The precision of the suggested work with improved features &CMCOA is 15.5 %, and 14.42 % superior to the proposed work without existing features & CMCOA and proposed work with existing features & CMCOA approaches. © 2022 Elsevier Ltd