Multi-dimensional deep learning drives efficient discovery of novel neuroprotective peptides from walnut protein isolates.

Abstract

Neurodegenerative diseases, such as Alzheimer's and Parkinson's, are multi-factor induced neurological disorders that require management from multiple pathologies. The peptides from natural proteins with diverse physiological activity can be candidates as multifunctional neuroprotective agents. However, traditional methods for screening neuroprotective peptides are not only time-consuming and laborious but also poorly accurate, which makes it difficult to effectively obtain the needed peptides. In this case, a multi-dimensional deep learning model called MiCNN-LSTM was proposed to screen for multifunctional neuroprotective peptides. Compared to other multi-dimensional algorithms, MiCNN-LSTM reached a higher accuracy value of 0.850. The MiCNN-LSTM was used to acquire candidate peptides from walnut protein hydrolysis. Following molecular docking, behavioral and biochemical index experimental validation eventually found 4 hexapeptides (EYVTLK, VFPTER, EPEVLR and ELEWER) demonstrating excellent multifunctional neuroprotective properties. Therein, EPEVLR performed the best and can be investigated in depth as a multifunctional neuroprotective agent. This strategy will greatly improve the efficiency of screening multifunctional bioactive peptides, and it will be beneficial for the development of food functional peptides.