Bond strength between receptor binding domain of spike protein and human angiotensin converting enzyme-2 using machine learning

Abstract

The spike protein (S-protein) of SARS-CoV-2 plays an important role in binding, fusion, and host entry. In this study, we have predicted interatomic bond strength between receptor binding domain (RBD) and angiotensin converting enzyme-2 (ACE2) using machine learning (ML), that matches with expensive ab initio calculation result. We collected bond order result from ab initio calculations. We selected a total of 18 variables such as bond type, bond length, elements and their coordinates, and others, to train ML models. We then trained five well-known regression models, namely, Decision Tree regression, KNN Regression, XGBoost, Lasso Regression, and Ridge Regression. We tested these models on two different datasets, namely, Wild type (WT) and Omicron variant (OV). In the first setting, we used 90% of each dataset for training and 10% for testing to predict the bond order. XGBoost model outperformed all the other models in the prediction of the WT dataset. It achieved an R2 Score of 0.997. XGBoost also outperformed all the other models with an R2 score of 0.9998 in the prediction of the OV dataset. In the second setting, we trained all the models on the WT (or OV) dataset and predicted the bond order on the OV (or WT) dataset. Interestingly, Decision Tree outperformed all the other models in both cases. It achieved an R2 score of 0.997.