Abstarct: In this study, quantitative structure-property (QSPR) models were developed to predict the retention time of 107 blood-adsorbed compounds in antiviral granules. Molecular structures were drawn and optimized with Hypercom software. Then, 3224 molecular descxriptors were calculated with Dragon software. Four stepwise regression (SW) methods, particle swarm optimization (PSO) algorithm, firefly algorithm (FF) and genetic algorithm (GA) were used to select the optimal descxriptors. Modeling was performed with artificial neural network (ANN). The results showed that the PSO-ANN model had the best performance with a coefficient of determination of 0.9790 in the test set and a mean square error of 1.68. The SW-ANN, GA-ANN and FF-ANN models were ranked next with coefficients of determination of 0.9708, 0.9319 and 0.8992, respectively. The Y-randomization test confirmed the absence of random correlation. Other methods such as domain of application, regression plots and statistical parameters were also used to evaluate the developed models. The selected descxriptors in the superior model were mainly from topological, electronic and spatial groups. In general, the combination of PSO algorithm with artificial neural network provides an efficient method for predicting the retention time of absorbed compounds in blood and can be used as a reliable tool in the quality control of pharmaceutical products.