Search for: All records

The combination of machine learning (ML) models with chemistry-related tasks requires the description of molecular structures in a machine-readable way. The nature of these so-called molecular descriptors has a direct and major impact on the performance of ML models and remains an open problem in the field. Structural descriptors like SMILES strings or molecular graphs lack size-independence and can be memory intensive. Machine-learned descriptors can be of low dimensionality and constant size but lack physical significance and human interpretability. Sigma profiles, which are unnormalized histograms of the surface charge distributions of solvated molecules, combine physical significance with low dimensionality and size-independence, making them a suitable candidate for a universal molecular descriptor. However, their widespread adoption in ML applications requires open access to sigma profile generation, which is currently not available. This work details the development of OpenSPGen – an open-source tool for generating sigma profiles. Also presented are studies on the effect of different settings on the efficacy of the generated sigma profiles at predicting thermophysical material properties when used as inputs to a Gaussian process as a simple surrogate ML model. We find that a higher level of theory does not translate to more accurate results. We also provide further recommendations for sigma profile calculation and use in ML models. 

Group contribution (GC) models are powerful, simple, and popular methods for property prediction. However, the most accessible and computationally efficient GC methods, like the Joback and Reid (JR) GC models, often exhibit severe systematic bias. Furthermore, most GC methods do not have uncertainty estimates associated with their predictions. The present work develops a hybrid method for property prediction that integrates GC models with Gaussian process (GP) regression. Predictions from the JR GC method, along with the molecular weight, are used as input features to the GP models, which learn and correct the systematic biases in the GC predictions, resulting in highly accurate property predictions with reliable uncertainty estimates. The method was applied to six properties: normal boiling temperature (Tb), enthalpy of vaporization at Tb (ΔHvap), normal melting temperature (Tm), critical pressure (Pc), critical molar volume (Vc), and critical temperature (Tc). The CRC Handbook of Chemistry and Physics was used as the primary source of experimental data. The final collected experimental data ranged from 485 molecules for ΔHvap to 5640 for Tm. The proposed GCGP method significantly improved property prediction accuracy compared to the GC-only method. The coefficient of determination (R2) values of the testing set predictions are ≥0.85 for five out of six and ≥0.90 for four out of six properties modeled, and compare favorably with other methods in the literature. Tm was used to demonstrate one way the GCGP method can be tuned for even better predictive accuracy. The GCGP method provides reliable uncertainty estimates and computational efficiency for making new predictions. The GCGP method proved robust to variations in GP model architecture and kernel choice.