COSMO-RS based predictions for the SAMPL6 logP challenge

• Published in: Journal of computer-aided molecular design Vol. 34; no. 4; pp. 385 - 392
• Main Authors: Loschen, Christoph, Reinisch, Jens, Klamt, Andreas
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.04.2020; Springer Nature B.V
• Subjects: Algorithms; Animal Anatomy; Chemistry; Chemistry and Materials Science; Computer Applications in Chemistry; Histology; Liquid phases; Machine learning; Morphology; Parameterization; Physical Chemistry; COSMOquick; Liquid phase thermodynamics; Machine learning; COSMOtherm; COSMO-RS; Octanol–water partition coefficients; logP
• ISSN: 0920-654X; 1573-4951
• DOI: 10.1007/s10822-019-00259-z

Within the framework of the 6th physical property blind challenge (SAMPL6) the authors have participated in predicting the octanol–water partition coefficients (logP) for several small drug like molecules. Those logP values where experimentally known by the organizers but only revealed after the submissions of the predictions. Two different sets of predictions were submitted by the authors, both based on the COSMOtherm implementation of COSMO-RS theory. COSMOtherm predictions using the FINE parametrization level (hmz0n) obtained the highest accuracy among all submissions as measured by the root mean squared error. COSMOquick predictions using a fast algorithm to estimate σ-profiles and an a posterio machine learning correction on top of the COSMOtherm results (3vqbi) scored 3rd out of 91 submissions. Both results underline the high quality of COSMO-RS derived molecular free energies in solution.