An accurate and efficient procedure was developed for performing C NMR chemical shift calculations employing density functional theory with the gauge invariant atomic orbitals (DFT-GIAO). Benchmarking analysis was carried out, incorporating several density functionals and basis sets commonly used for prediction ofC NMR chemical shifts, from which the B3LYP/cc-pVDZ level of theory was found to provide accurate results at low computational cost. Statistical analyses from a large data set of C NMR chemical shifts in DMSO are presented with TMS as the calculated reference and with empirical scaling parameters obtained from a linear regression analysis. Systematic errors were observed locally for key functional groups and carbon types, and correction factors were determined. The application of this process and associated correction factors enabled assignment of the correct structures of therapeutically relevant compounds in cases where experimental data yielded inconclusive or ambiguous results. Overall, the use of B3LYP/cc-pVDZ with linear scaling and correction terms affords a powerful and efficient tool for structure elucidation.

The calculation of N NMR chemical shifts has been systematically investigated using density functional theory-gauge including/invariant atomic orbitals (DFT-GIAO) approximation at the B3LYP/cc-pVDZ level of theory. General linear regression terms forN chemical shift predictions were calculated for nitromethane and liquid ammonia references in DMSO. Both aliphatic and aromatic nitrogens were studied using a diverse set of molecular scaffolds. Statistical error analysis between experiment and prediction revealed that, with the exception of primary amines, 95% of linear scaled N-15 chemical shifts are within a ±9.56 ppm range. Comparison of the N calculated isotropic chemical shifts with the experimentally determined chemical shifts provided accurate assignment of the correct structure in cases where experimental data was ambiguous or inconclusive. Application ofN prediction proved to be highly effective in identifying the correct regio-isomer, oxidation state, protonation state and preferred tautomer in solution.

Methods to predict crystallization behavior for active pharmaceutical ingredients (APIs) can serve as an important guide in small molecule pharmaceutical development. Here, we describe solvate formation propensity prediction for pharmaceutical molecules via a machine learning approach. Random forests (RF) and support vector machine (SVM) algorithms were trained and tested with data sets extracted from Cambridge Structural Database (CSD). The machine learning models, requiring only 2D structures as input, were able to predict solvate formation propensity for organic molecules with up to 86% success rate. Performance of the models was demonstrated with a collection of 20 pharmaceutical molecules.

NMR chemical shift prediction at the B3LYP/cc-pVDZ level of theory was used to develop a highly accurate probability theory algorithm for the determination of the stereochemistry of diastereomers as well as the regiochemistry. DFT-GIAO calculations were performed for each conformer using geometry optimization and a CPCM solvent model. Boltzmann averaged shielding constants were converted to chemical shifts for H andC, using the generalized linear scaling terms determined in four different solvents for H andC and extended to N in DMSO. The probability theory algorithm, D iCE, was based on the DP4 method and developed forH, C, andN NMR using individual and combined probability data. The chemical shift calculation errors were fitted to a Student's t-distribution for H andC and a normal distribution for N. The application yielded a high accuracy for structural assignment with a low computational cost.

Minimalist secondary structure mimics are typically made to resemble one interface in a protein-protein interaction (PPI), and thus perturb it. We recently proposed suitable chemotypes can be matched with interface regions directly, without regard for secondary structures. This communication describes a modular synthesis of a new chemotype 1, simulation of its solution-state conformational ensemble, and correlation of that with ideal secondary structures and real interface regions in PPIs. Scaffold 1 presents amino acid side-chains that are quite separated from each other, in orientations that closely resemble ideal sheet or helical structures, similar non-ideal structures at PPI interfaces, and regions of other PPI interfaces where the mimic conformation does not resemble any secondary structure. Sixty-eight different PPIs where conformations of 1 matched well were identified. A new method is also presented to determine the relevance of a minimalist mimic crystal structure to its solution conformations. Thus DLD-1faf crystallized in a conformation that is estimated to be 0.91 kcal•mol−1 above the minimum energy solution state.