Pre-launch insight to Torx™, a new platform for small molecule discovery chemistry
Read full press release
Leverkusen, Germany – 30th March 2016 – COSMOlogic, provider of software solutions based on quantum chemistry and thermodynamics, has performed best in this year’s Drug Design Data Resource (D3R) SAMPL challenge with their software package COSMOtherm for the prediction of thermodynamic properties of liquids.
Each year, SAMPL presents the molecular modeling community with blind challenges designed to test the accuracy of their techniques. COSMOtherm performed best in the challenge designed to test the prediction of distribution coefficients of drug-like molecules partitioning between water and extremely non-polar organic solvents (such as cyclohexane). These predictions are important for the prediction of drug de-solvation and the penetration of drugs through cell membranes.
“We are delighted that COSMOtherm proved its accuracy by coming out a clear first in this highly contested challenge,” says Prof Andreas Klamt, CEO of COSMOlogic. “This year’s challenge included the blind prediction of the water-cyclohexane partitioning on 53 highly demanding drug-like molecules. COSMOtherm is widely accepted as the industry standard for the thermodynamic properties of liquids, and these exceptionally good results are a further endorsement of our methods.”
The official presentation of the results during the 1st Annual 3DR Workshop (9-11 Mar, 2016, La Jolla) revealed that the COSMOtherm prediction performed best in overall RMS-deviation to experiment and also in rank ordering the molecules. The COSMOtherm prediction yielded a correlation coefficient of 0.85 and an RMS error of 2.1 log-units. The second best contribution achieved a correlation of 0.75 but an RMS of 2.7 log-units. Further discussion between organizers and participants lead to the conclusion that a significant proportion of the errors was likely to be due to problems arising from the physical measurement of partitioning of very polar compounds in alkanes, i.e. there is very significant noise in the experimental data which then leads to higher errors between predicted and experimental data.