Background: Functional cure for Hepatitis B virus (HBV) by inhibiting HBV surface antigen (HBsAg) is crucial. We aimed to develop a predictive quantitative structure-activity relationship (QSAR) model on a ligand-based pharmacophore (LBP) derived from already known HBsAg secretion inhibitors in the present study.Methods: A LBP model was developed using active HBsAg secretion inhibitors as both trainings- and test-sets using LigandScout v3.12 software. The best model with the highest score was used for high throughput screening (HTS) screening of a virtual library comprising 720,000 compounds. A QSAR model was developed by a stepwise multiple linear regression (MLR) on ~2700 descriptors with a confidence interval (CI) of 95%. The test set validated the QSAR model. The goodness of fit statistics evaluated the fitness of the model. A comparable R2 and adjusted R2 were considered as the lack of overfitting. Further RMSE and Q2 statistics were measured for testing the model on the validation set. Principal component analysis (PCA) was also evaluated to estimate the predictor variables' associations and impact on the model.Results: 34 active anti-HBsAg compounds were used to develop an LBP model. 9/34 of compounds with higher clustering pharmacophore-fit scores were tagged as the training set, and the rest of the inhibitors were used as the test set. The best model had a 0.8832 fit score. HTS resulted in 10 potential hit compounds with a fit score of 101.44±0.65. A QSAR model was developed with two response variables, including Yindex and GATS8m, with substantial variance information (p < 0.05). The model was well fitted (R2 = 0.9563, MSE = 0.0023). The model was not predictive on the test set (Q2 = 0.00, RMSE = 0.8153). The PCA results of two factors demonstrated a substantial variance data of both predictor variables. Conclusion: The present study showed a reliable pharmacophore modeling based on known active inhibitors of HBsAg and a well-fitted predictive QSAR model on the LBP. The model can be applied to the chemical libraries fitted to the LBP model, and the QSAR equation would estimate the biological activities of the hit compounds with 95.63% accuracy with only two Yindex and GATS8m descriptors.