A disease-related essential protein prediction model based on the transfer neural network

Abstract

Essential proteins play important roles in the development and survival of organisms whose mutations are proven to be the drivers of common internal diseases having higher prevalence rates. Due to high costs of traditional biological experiments, an improved Transfer Neural Network (TNN) was designed to extract raw features from multiple biological information of proteins first, and then, based on the newly-constructed Transfer Neural Network, a novel computational model called TNNM was designed to infer essential proteins in this paper. Different from traditional Markov chain, since Transfer Neural Network adopted the gradient descent algorithm to automatically obtain the transition probability matrix, the prediction accuracy of TNNM was greatly improved. Moreover, additional antecedent memory coefficient and bias term were introduced in Transfer Neural Network, which further enhanced both the robustness and the non-linear expression ability of TNNM as well. Finally, in order to evaluate the identification performance of TNNM, intensive experiments have been executed based on two well-known public databases separately, and experimental results show that TNNM can achieve better performance than representative state-of-the-art prediction models in terms of both predictive accuracies and decline rate of accuracies. Therefore, TNNM may play an important role in key protein prediction in the future.

Recent Advances in Deep Learning for Protein-Protein Interaction Analysis: A Comprehensive Review

“…A host of studies [ 109 , 110 , 111 , 112 , 113 ] have demonstrated the potential of these methodologies to enhance our understanding of PPI mechanisms and to develop predictive models with superior accuracy. A summary of recent research can be seen in Table 9 .…”

“…Among these, Chen et al [ 109 ] put forward TNNM, a transfer neural-network-based model specifically designed for the prediction of essential proteins. The researchers achieved this by extracting raw features from multiple biological data sources and demonstrating enhanced prediction performance compared to existing models.…”

“…Introduced HDIContact, a deep learning framework for inter-protein residue contact prediction, showcasing promising results for understanding PPI mechanisms. Among these, Chen et al [109] put forward TNNM, a transfer neural-networkbased model specifically designed for the prediction of essential proteins. The researchers achieved this by extracting raw features from multiple biological data sources and demonstrating enhanced prediction performance compared to existing models.…”