Machine learning transition temperatures from 2D structure

A priori knowledge of physicochemical properties such as melting and boiling could expedite materials discovery. However, theoretical modeling from first principles poses a challenge for efficient virtual screening potential candidates. As an alternative, the tools data science are becoming increasingly important exploring chemical datasets predicting material properties. Herein, we extend molecular representation, or set descriptors, developed quantitative structure-property relationship by Yalkowsky coworkers known Unified Physicochemical Property Estimation Relationships (UPPER). This representation has group-constitutive geometrical descriptors that map to enthalpy entropy; two thermodynamic quantities drive thermal phase transitions. We UPPER include additional information about sp 2 -bonded fragments. Additionally, instead using in series thermodynamically-inspired calculations, per Yalkowsky, use construct vector with machine learning techniques. The concise easy-to-compute combined gradient-boosting decision tree model, provides appealing framework experimental transition temperatures diverse space. An application energetic shows method is predictive, despite relatively modest energetics reference dataset. also report competitive results on public points ( i.e. , OCHEM, Enamine, Bradley, Bergström) comprised over 47k structures. Open source software available at https://github.com/USArmyResearchLab/ARL-UPPER . • ARL-UPPER encodes enthalpic entropic contributions Leveraging XGBoost builds accurate point predictors. ARL-UPPER’s strong performance validated chemically datasets. Easy-to-use code