STX150: A Curated Database for the Development and Validation of DP4+ Methods

Abstract Structural elucidation remains a central challenge in modern chemistry, particularly in the determination of relative stereochemistry when experimental spectroscopic data are insufficient. Among computational approaches, the DP4+ method has become a widely adopted tool for NMR-based stereochemical assignment by combining experimental and calculated chemical shifts within a probabilistic framework. However, the computational cost associated with geometry optimization (typically performed at the DFT level) remains a major limitation for large data sets and high-throughput studies. Herein, we introduce STX150, a curated database of 150 structurally diverse yet conformationally simple molecules designed to facilitate rapid benchmarking and statistical parametrization within the DP4+ framework. This data set preserves the statistical robustness of the original training sets while reducing the computational cost of the parametrization procedure by approximately 98%, enabling efficient exploration of theoretical levels. Using an automated workflow integrated with STX150, we systematically evaluated multiple density functionals and basis sets for NMR chemical shift prediction and analyzed the impact of different geometry optimization strategies. In particular, the performance of GFN-xTB geometries was examined as an alternative to conventional molecular mechanics and DFT optimizations. The results show that xTB provides significantly improved structural descriptions compared with MMFF, leading to correct DP4+ assignments in up to ∼91% of the cases in a benchmark set of 110 molecules, while maintaining a much lower computational cost than DFT geometries. At the highest level of theory, the conventional PCM/mPW1PW91/6–31+G**//B3LYP/6–31G* protocol achieved perfect assignment performance for the entire data set. Finally, the xTB-DP4+ protocol has been implemented in the open-access DP4+App software, enabling automated structural assignments using semiempirical geometries. Together, the STX150 data set, automated benchmarking workflow, and xTB-enabled protocols provide a scalable and efficient platform for rapid and reliable NMR-based structural elucidation.