Density Functional Theory (DFT) is the most important foundation in computational quantum chemistry. DFT provides a solid theoretical foundation for predicting the physical and chemical properties of molecules in materials. Although quantum computing methods based on DFT give very accurate results, they require huge computational resources. Therefore, high-throughput applications, which require the calculation of material properties for a large number (tens or hundreds of thousands) of molecules, cannot utilize DFT. Recent advances in deep learning, especially message-passing neural networks (MPNNs), offer a promising alternative for large-scale molecular property predictions. This project aims to evaluate the accuracy and effectiveness of the MPNNs deep learning model compared to DFT in predicting molecular properties. We use tests like R-Squared (0.87÷0.99), Pearson's R (0.97÷0.99), and Spearman's Rho (0.91÷0.99) to look at the predictions of 12 physicochemical properties for about 13,000 molecules in the test set. This shows that the potential for applying MPNNs to building a database system of new and doped materials is huge in the future.