Abstract
Mining of anti-diabetic dipeptidyl peptidase IV (DPP-IV) inhibitory peptides (DPP-IV-IPs) is currently a costly and laborious process. Due to the absence of rational peptide design rules, it relies on cumbersome screening of unknown enzyme hydrolysates. Here, we present an enhanced deep learning (DL) model called BERT-DPPIV, specifically designed to classify DPP-IV-IPs and exploring their design rules to discover potent candidates. The end-to-end model utilizes a fine-tuned bidirectional encoder representations (BERT) architecture to extract structural/functional information from input peptides and accurately identify DPP-IV-Ips from input peptides. Experimental results in benchmark dataset showed BERT-DPPIV yielded state-of-the-art accuracy of 0.894, surpassing the 0.797 obtained by sequence-feature model. Furthermore, we leverage the attention mechanism to uncover that our model could recognize restriction enzyme cutting site and specific residues that contribute to the inhibition of DPP-IV. Moreover, guided by BERT-DPPIV, proposed design rules of DPP-IV inhibitory tripeptides and pentapeptides were validated and they can be used to screen potent DPP-IV-IPs.
All Related Versions

This publication has 29 references indexed in Scilit: