Concentration-Dependent bidirectional regulation of adenosine receptor A1 explored through machine learning

Abstract

ObjectiveThis study aims to predict the IC50 of adenosine receptor A1 agonists using machine learning methods, demonstrating the concentration-dependent bidirectional regulatory effect of this drug target, and to explore the relationship between this effect and the properties and structure of the drug molecules. MethodsAn IC50 and EC50 database for adenosine receptor A1 was established, and a differential analysis of molecular weight, hydrogen bond donors/acceptors, and LogP of these compounds was conducted. Four machine learning models, including Random Forest, Support Vector Machine, Gradient Boosting, and Fully Connected Neural Network, were trained on the IC50 dataset to develop predictive models, which were evaluated using MSE and R2 score. The best-performing predictive model was then used to predict the IC50 on the EC50 dataset and compared with the actual EC50. A 95% confidence interval for the predicted IC50 was obtained to evaluate the prediction results. Drugs with different orders of inhibitory/agonistic effects were set as two datasets, and T-tests and principal component analysis were used to analyze differences in molecular weight, hydrogen bond donors/acceptors, LogP, and topological structure between the two datasets to identify the molecular structural differences causing the inconsistency in the order of inhibitory/agonistic effects. ResultsA database containing 748 adenosine receptor A1 inhibitors and 303 agonists was successfully constructed. The Random Forest model performed best in the machine learning model training. The R2 score, MSE, RMSE, MAE, and MAPE values are 0.84, 7307.18, 85.48, 46.34, and 7.46, respectively, indicating a good dispersion of the regression curve. The prediction results showed that most agonists also have inhibitory effects. The analysis of differences in molecular weight and hydrogen bond donors/acceptors showed that the number of hydrogen bond donors and LogP had highly significant differences between the two groups (P < 0.001). PCA of topological structures showed that while there were some similarities in the compound structures of the two groups, there were also clear differences between the two clusters. ConclusionsThis study has found that drugs acting on the adenosine receptor A1 demonstrate concentration-dependent bidirectional regulatory effects due to differences in structure, the number of hydrogen bond donors, and LogP.