Prediction of ADME Properties

Poor intestinal absorption of a potential drug molecule can be related to poor physicochemical properties and/or poor membrane permeation. Poor membrane permeation could be due to low paracellular or transcellular permeability or the net result of efflux from transporter proteins including MDRl (P-gp) or MRP proteins situated in the intestinal membrane. Cell lines with only one single efflux transporter are currently engineered for in vitro permeability assays to get suitable data for reliable QSAR models. In addition, efforts to gain deeper insight into P-gp and ABC on a structural basis are going on [131, 132]. 

Discrimination of efflux, active or passive transport is already feasible by suitable in vitro experiments. For instance, the PAMPA assay detects passive transport only, while Caco-2 cells include transporters. A comparison between transport in PAMPA and Caco-2 cells by a calibration plot reveals compounds with greater or less transport in Caco-2 cells than in PAMPA. These compounds should be tested in uptake and efflux transport assays in order to gain deeper insight into absorption fate. 

Several in silica models for prediction of oral absorption are available [133-136]. Simple models are based on only few descriptors like logP, logD, or polar surface area (PSA), while they are only applicable if the compounds are passively absorbed. In case of absorption via active transporters or if efflux is involved, prediction of absorption is still not successful. 

GastroPlus [137] and IDEA [138] are absorption-simulation models based on in vitro input data like solubility, Caco-2 permeability and others. They are based on advanced compartmental absorption and transit (ACAT) models in which physicochemical concepts are incorporated. Both approaches were recently compared and are shown to be suitable to predict the rate and extent of human absorption [139]. 

Prediction of bioavailability from molecular structure is quite difficult, since bioavailability depends on absorption and first-pass clearance [141]. By applying fuzzy adaptive least squares , Yoshida and Topliss generated a QSAR model using logD at pH 7.4 and 6.5 as input for physicochemical properties and the presence/absence of certain functional groups as structural input. They achieved a classification of drugs into one of four bioavailability categories with an overall accuracy of 60% [142]. 

---

Prediction of ADME properties should be simple, since the number of descriptors underlying the properties is relatively small, compared to the number associated with effective drug-receptor binding space. In fact, prediction of ADME is difficult The current ADME experimental data reflect a multiplicity of mechanisms, making prediction uncertain. Screening systems for biological activity are typically single mechanisms, where computational models are easier to develop [1],... 

Computational Prediction of Drug-Likeness 29 Table 2.1 Machine learning algorithms used for the prediction of ADME properties. 

Updating and validating such algorithms and their databases are also critical aspects. At this time, the European Cooperation in the Field of Scientific and Technical Research (Project COST B15) had begun an independent evaluation of existing expert systems used in the in silico prediction of ADME properties, with a view of publishing a consensus paper. 

A.P. Beresford, M. Segall, M.H. Tarbit, In silico prediction of ADME properties are we making progress Curr. Opin. Drug Discov. Devel. 2004, 7, 36-42. 

The partition coefficient and aqueous solubility are properties important for the study of the adsorption, distribution, metabolism, excretion, and toxicity (ADME-Tox) of drugs. The prediction of the ADME-Tox properties of drug candidates has recently attracted much interest because these properties account for the failure of about 60 % of all drug candidates in the clinical phases. The prediction of these properties in an early phase of the drug development process could therefore lead to significant savings in research and development costs. 

Although prediction of ADME/PK in man may be the primary purpose for the pre-clinical studies, it is also important that potential new drugs have acceptable properties in toxicology species. Without these it can be very difficult to generate adequate safety margins to allow studies in man to start. It is also likely that the development safety assessment program will be difficult and hence slow. 

Castillo-Garit, J.A., Marrero-Ponce, Y., Torrens, F., Garcia-Domenech, R. Estimation of ADME properties in drug discovery predicting Caco-2 cell permeability using atom-based stochastic and non-stochastic linear indices. J. Pharm. Sci. 2008, 97, 1946-76. 

In silico techniques have gained wide acceptance as a tool to support the drug discovery and optimization process. Binding mode predictions via docking, affinity predictions via QSAR and CoMFA, or the prediction of ADME(T) properties are routinely applied [1-3]. 

This is the framework where computational chemistry could play an important role in the prediction of these properties in order to obtain more efficient and faster drug discovery cycles. To obtain useful descriptors for ADME properties is not an easy task. A large number of descriptors have been developed [4], all of which have major limitations in terms of relevance, interpretability or speed of calculation. 

VolSurfwas initially validated on oral absorption [16, 17] and blood-brain-barrier permeation [18] models (see belovi ). VolSurf has continued to be developed to improve in silico predictions for ADME properties, although its use has also been extended to receptor-based evaluation of binding affinity [19, 20], While other soft-ivare tools for ADME modeling are available (see, e.g., [21]), the MIF-based collection of sofhvare and models available from Molecular Discovery (MD) is both extensive and ivell validated by the private sector. Three programs from MD, VolSurf, MetaSite and Almond, are particularly suited for rapid evaluation of large compound sets [22] in connection ivith ADME/Tox related properties ... 

This perspective has examined the approaches to molecular modeling and drug design and emphasized their limitations. The reader should be aware, however, that these tools are daily used on many problems of therapeutic interest with increasing success. This is clearly witnessed by publications of such studies in almost every issue of current major journals. For specific application areas, such as RNA (490, 491), DNA (492-496), membrane (497-507), or peptidomimetic modeling (382, 508-513), the reader is referred to the literature. The prediction of molecular properties, such as log P and correlation between substructures and metabolism, has led to a dramatic increase in efforts to correlate adsorption, distribution (514), metabolism (515-617), and elimination (ADME) with chemical... 

In spite of the high interest in the parameters and in spite of the fact that more and more software companies offer products to calculate them, reliable calculation of ADME properties remains difficult The results are only acceptable if the molecular structures of the training set (for which experimental data are known) and the structures of interest are similar. Typically, the methods achieve a high degree of predictivity (> 90% correct classification) within the training data. Nevertheless, most software packages fail completely in realistic tests [47]. 

Model systems needed for particular studies are listed in Figure 5.2. In addition, in silico models are developed to predict drug behaviors based on physicochemical properties of drugs or drug candidates, crystal structures of a protein (an enzyme or a transporter), and database of ADME properties generated in laboratories. Therefore, experimental models are important for ADME studies. The objective of this chapter is to discuss strategy and applications of experimental models in drug metabolism and disposition. 

---