Drug-target interaction prediction in coronavirus disease 2019 case using deep semi-supervised learning model

Abstract

Coronavirus disease 2019 (COVID-19) is an infectious disease of the respiratory system that caused a pandemic in 2020. There is still not any effective special treatment to cure it. Drug repositioning is used to find an effective drug for curing new diseases by finding new efficacy of registered drug. The new efficacy can be conducted by elaborating the interactions between compounds and proteins (DTI). Deep Semi-Supervised Learning (DSSL) is used to overcome the lack of DTI information. DSSL utilizes unsupervised learning algorithms such as Stacked Auto Encoder (SAE) as pre-training for initializing weights on the Deep Neural Network (DNN). This study uses DSSL with a feature-based chemogenomics approach on the data resulted from the exploration of potential anti-coronavirus treatment. This study finds that the use of fingerprints for compound features and Dipeptide Composition (DC) for protein features gives the best results on accuracy (0.94), recall (0.83), precision (0.817), F-measure (0.822), and AUROC (0.97). From the test data predictions, 1766 and 929 positive interactions are found on the test data and herbal compounds, respectively.