Sequence-Based Prediction of Food-Originated ACE Inhibitory Peptides Using Deep Learning Algorithm

In recent years, food originated bioactive peptides became a promising source of potential therapeutic agents. Predicting the biological activity of these peptides is crucial for the discovery and development of functional foods and effective peptides-based drugs. Antihypertensive peptides (AHTPs) are certainly the most reported food-derived peptides. These peptides inhibit a key enzyme in renin-angiotensin system, named angiotensin-converting enzyme (ACE), resulting in lowering of blood pressure. So far, AHTPs are obtained mainly by in vitro and in vivo protocols. This is a rather expensive and time-consuming procedure, and often require months of hard work, which is not always successful. To overcome this shortcoming, machine learning (ML) approaches are increasingly used. In this study, a Long Short Term Memory (LSTM) is used for prediction of food-derived ACE inhibitory peptides. It was chosen this recurrent deep neural network as the most suitable for sequence-based prediction. The positive datasets are collected from the following food-derived peptide databases AHTPDB, FeptideDB, BIOPEP-UWM and BioPepDB, while the negative ones peptides without antihypertensive function were gathered. Then, the feature descriptors are generated via the Chou's pseudo amino acid composition method. They are inputted to the deep neural network classifier. Finally, the proposed LSTM approach is compared with Random Forest (RF) and Support Vector Macines (SVM) classifiers. It was demonstrated by 5-fold cross-validation that the deep learning algorithm has higher predictive accuracy than the other ML algorithms. This makes it suitable for identification of AHTPs.