Protein-Based Data Augmentation for the Prediction of Peptide Toxicity Using Deep Learning

Peptides have a promising pharmaceutical value with its small side effect and high specificity. While their unclear toxicity is one of the key bottlenecks preventing them from being widely used in clinical practice. To save time and labor, many computation-Aided models have been proposed to do binary classification of peptide toxicity. However, limited by the availability of datasets about peptide toxicity, it is hard to improve the performance of computational aided models. Given the situation that there are a substantial number of available protein toxicity data, we proposed a simple deep learning model with convolution layer and LSTM and applied protein-based data augmentation on it. Experimental results show there is an obvious increase in precision using protein-based data augmentation on the proposed deep learning model.