Three different sets of molecules were used to evaluate COSMOtherm predictions of water solubility. The first and second sets are each comprised of over 1000 compounds of all differents kinds, while the third has about 200, all of them drugs. Not all of these compounds can be found in the COSMOtherm database, so COSMOfrag was used to predict the solubility values. All three resulting graphs are shown below.
In the first two, the large majority of compounds show a resonable deviation from the experimental value, while a few outliers fall significantly farther. However, many of these outliers seem questionable, and are therefore not included in the standard deviation calculation. They are discussed more in depth in the next page.
Standard Deviation: 0.85
The set's name is alogPs train and valid
Standard Deviation: 0.98
The set's name is ESOL
Standard Deviation: 1.54
Study conducted by Antoine Hervier
Go to the next page
back to various examples