文摘
Unfavorable lipophilicity and water solubility cause many drug failures; therefore theseproperties have to be taken into account early on in lead discovery. Commercial tools forpredicting lipophilicity usually have been trained on small and neutral molecules, and are thusoften unable to accurately predict in-house data. Using a modern Bayesian machine learningalgorithm-a Gaussian process model-this study constructs a log D7 model based on 14556drug discovery compounds of Bayer Schering Pharma. Performance is compared with supportvector machines, decision trees, ridge regression, and four commercial tools. In a blind test on7013 new measurements from the last months (including compounds from new projects) 81%were predicted correctly within 1 log unit, compared to only 44% achieved by commercialsoftware. Additional evaluations using public data are presented. We consider error bars foreach method (model based error bars, ensemble based, and distance based approaches), andinvestigate how well they quantify the domain of applicability of each model.