for Physical Chemists by Physical Chemists

Home Up Contents

Toni Llinās
Christine Prosser Paula Garcia John Comer Shenaz Nunhuck Darren Edwards Toni Llinās


ADMET prediction: fiction or reality?

Toni Llinās, Unilever Centre for Molecular Informatics, Pfizer institute for Pharmaceutical Materials Sciences, University of Cambridge

Reviews of the performance of ADMET prediction methods and models often end with the statement that is "relatively simple to develop models that fit the entire data set, but that, typically, such models do not predict new data sets well", what is needed, all reviewers agree, is more and better data. It doesn't matter how good a new model is, because in the end a model is as good as the data it is based on.

A criteria-based evaluation of, aqueous solubility (Sw) and octanol-water partition coefficient (Kow) data sources, for example, shows that 95-100 percent of the database literature is of poor or unevaluable quality. The accuracy and reliability of the vast majority of the data are unknown due to inadequate documentation of the methods of determination used by the authors (for example, estimates of precision have been reported for only 20% of experimental Sw data).

We need a reliable and accurate determination of solubility values. The problem when measuring aqueous solubility, either kinetic or thermodynamic, is to define the conditions used. At present, when we read a solubility value from a paper, most of the times we do not know how this value has been obtained. To be able to compare, and therefore to build a reliable solubility data base, we should specify exactly which solubility we are talking about and how it was measured, and, ideally, scientists all around the world should use the same "standard" conditions and notations, i.e. we need to know if the compound was just thrown into water or if it was an aqueous buffered media, which buffer, what pH, constant ionic strength, temperature and if it really reached equilibrium or it was just supposed. We need to know what kind of filter was used (if any), and the temperature at which we perform this filtration is important. An analysis of the purity of the starting material is crucial; no solubility data should be given if the compound is not at least 99.5 % pure. A crystallographic study could be made on our initial compound before starting the measurements, and, if possible, after, to be sure that the reported solubility corresponds to the same crystal structure and not to a polymorph, and last, we need to assess the quality of the data repeating the experiment several times, giving statistical analysis together with the final values (estimates of precision, errors, deviations…)

Contact Information

For more information, please send an email to

You are visitor number

Send mail to with questions or comments about this web site.
Last modified: 28 April 2008