Predicting peptide bioactivity with long short-term memory: a compact modelfor sequence-based classification

Bioactive peptides (BPs) are short-chain amino acids that have been shown to have several health benefits. Laboratory-based identification of BPs is a time-consuming and labor-intensive process, hence multiple in-silico methods have been developed and used to determine BPs from various sources. In this study, a model, based on a long short-term memory (LSTM) binary classification model, was created and trained on existing bioactive peptides from literature and available databases to determine if an input peptide is bioactive or not. Peptide sequences used were from plant and animal sources. In addition, a set of non-bioactive peptides was used as the null set. The study used word encoding on each sequence wherein each sequence was preprocessed to have spaces in between each amino acid before being tokenized. Results showed an 89.13% validation Accuracy and a 0.8174 f-measure value for the model that utilized the dataset containing all peptides used. Model performance can be improved by adding to the dataset and trying different encoding methods, which retain more information. The constructed model and its prediction can be used as a baseline for further studies in the predictive determination of bioactive peptide sequences from food products.