Identification of essential proteins based on Local Random Walk and Adaptive Multi-View Multi-Label Learning

Accumulating evidences have indicated that essential proteins play vital roles in human physiological process. In recent years, although researches on prediction of essential proteins are developing rapidly, they suffer from various limitations including unsatisfactory data suitability and low accuracy of predictive results. In this manuscript, a novel method called RWAMVL was proposed to predict essential proteins based on Random Walk and Adaptive Multi-View multi-label Learning. In RWAMVL, taking into account that the inherent noise is ubiquitous in existing datasets of known protein-protein interactions (PPIs), a variety of different features including biological features of proteins and topological features of PPI networks would be obtained by adopting adaptive multi-view multi-label learning first. And then, an improved random walk method would be designed to detect essential proteins based on these different features. Finally, in order to accurately verify the predictive performance of RWAMVL, intensive experiments would be done to compare RWAMVL with multiple state-of-the-art predictive methods under different expeditionary frameworks, and comparative results illustrated that RWAMVL could achieve high prediction accuracy than all these competitive methods as a whole, which demonstrated that RWAMVL may be a potential tool for prediction of key proteins in the future.