Generation of in-silico cytochrome P450 1A2, 2C9, 2C19, 2D6, and 3A4 inhibition QSAR models.

In-silico models were generated to predict the extent of inhibition of cytochrome P450 isoenzymes using a set of relatively interpretable descriptors in conjunction with partial least squares (PLS) and regression trees (RT). The former was chosen due to the conservative nature of the resultant models built and the latter to more effectively account for any non-linearity between dependent and independent variables. All models are statistically significant and agree with the known SAR and they could be used as a guide to P450 liability through a classification based on the continuous pIC50 prediction given by the model. A compound is classified as having either a high or low P450 liability if the predicted pIC(50) is at least one root mean square error (RMSE) from the high/low pIC(50) cut-off of 5. If predicted within an RMSE of the cut-off we cannot be confident a compound will be experimentally low or high so an indeterminate classification is given. Hybrid models using bulk descriptors and fragmental descriptors do significantly better in modeling CYP450 inhibition, than bulk property QSAR descriptors alone.

State-of-the-art tools for computational site of metabolism predictions: comparative analysis, mechanistical insights, and future applications.

In drug design, it is crucial to have reliable information on how a chemical entity behaves in the presence of metabolizing enzymes. This requires substantial experimental efforts. Consequently, being able to predict the likely site/s of metabolism in any compound, synthesized or virtual, would be highly beneficial and time efficient. In this work, six different methodologies for predictions of the site of metabolism (SOM) have been compared and validated using structurally diverse data sets of drug-like molecules with well-established metabolic pattern in CYP3A4, CYP2C9, or both. Three of the methods predict the SOM based on the ligand's chemical structure, two additional methods use structural information of the enzymes, and the sixth method combines structure and ligand similarity and reactivity. The SOM is correctly predicted in 50 to 90% of the cases, depending on method and enzyme, which is an encouraging rate. We also discuss the underlying mechanisms of cytochrome P450 metabolism in the light of the results from this comparison.

Development, interpretation and temporal evaluation of a global QSAR of hERG electrophysiology screening data.

A 'global' model of hERG K(+) channel was built to satisfy three basic criteria for QSAR models in drug discovery: (1) assessment of the applicability domain, (2) assuring that model decisions can be interpreted by medicinal chemists and (3) assessment of model performance after the model was built. A combination of D-optimal onion design and hierarchical partial least squares modelling was applied to construct a global model of hERG blockade in order to maximize the applicability domain of the model and to enhance its interpretability. Additionally, easily interpretable hERG specific fragment-based descriptors were developed. Model performance was monitored, throughout a time period of 15 months, after model implementation. It was found that after this time duration a greater proportion of molecules were outside the model's applicability domain and that these compounds had a markedly higher average prediction error than those from molecules within the model's applicability domain. The model's predictive performance deteriorated within 4 months after building, illustrating the necessity of regular updating of global models within a drug discovery environment.

Reaction site mapping of xenobiotic biotransformations.

Predictive metabolism methods can be used in drug discovery projects to enhance the understanding of structure-metabolism relationships. The present study uses data mining methods to exploit biotransformation data that have been recorded in the MDL Metabolite database. Reacting center fingerprints were derived from a comparison of substrates and their corresponding products listed in the database. This process yields two fingerprint databases: all atoms in all substrates and all reacting centers. The metabolic reaction data are then mined by submitting a new molecule and searching for fingerprint matches to every atom in the new molecule in both databases. An "occurrence ratio" is derived from the fingerprint matches between the submitted compound and the reacting center and substrate fingerprint databases. Normalization of the occurrence ratio within each submitted molecule enables the results of the search to be rank-ordered as a measure of the relative frequency of a reaction occurring at a specific site within the submitted molecule. Predictive performance that would allow this method to be used by drug discovery teams to generate useful hypotheses regarding structure metabolism relationships was observed.