Computational models for the prediction of intestinal membrane permeability

Abstract: Lead compounds generated in high-throughput drug discovery programs often have unfavorable biopharmaceutical properties, resulting in a low success rate for such drug candidates in clinical development. Efficient and reliable methods that predict biopharmaceutical properties, such as intestinal permeability and solubility are therefore required in order to reduce the attrition rate during development of these compounds.One aim of this thesis was to identify molecular properties that are important for intestinal drug permeability using a wide range of drugs and model compounds. A second aim was to develop computational models for predicting intestinal drug permeability based on these properties.The calculated molecular descriptors ranged from the simple counting of atoms and fragments to more complex descriptors derived from molecular mechanics and quantum mechanics calculations. Particular attention was given to descriptors associated with molecular surface areas. Descriptors calculated by the various methods were used to establish structure-permeability relationships for conventional drugs, peptide derivatives and large, lipophilic compounds generated by high-throughput pharmacological screening. Caco-2 cell monolayer permeabilities were determined for a structurally diverse set of compounds and were used to predict human intestinal membrane permeability and to develop computational models.From these investigations, several new models for the computational prediction of intestinal membrane permeability were developed. Models were developed that are suitable for the prediction of membrane permeability to specific types of drugs, as well as models that are more generally applicable. One of these general models is based on partitioned total molecular surface areas, and this model can be used to predict intestinal membrane permeability to structurally diverse compounds. It was also demonstrated how these models can be applied in a manner that increases both the accuracy of the prediction and the throughput. In addition, a simplified protocol based on Caco-2 cells for the experimental prediction of intestinal permeability was developed. These improvements can be used to construct highly effective experimental and computational filters for use in drug discovery and development.