Computational or virtual screening has received much attention in the last few years. In silico models which can accurately predict the membrane permeability of test drugs based on lipophilicity, hydrogen bonding capacity, molecular size, polar surface area, and quantum properties has the potential to specifically direct the chemical synthesis and therefore, revolutionize the drug discovery process. Such an in silico predictive model would minimize extremely time consuming steps of synthesis as well as experimental studies of thousands of test compounds.
Lipinski et al. (1997) proposed an in silico computational method for qualitatively predicting the developability of compounds. The "rules of five" proposed by them predicted lower permeabilities for compounds with more than five H-bond donors, ten H-bond acceptors, with molecular weight greater than 500 and c Log P > 5. Using this completely empirical model, useful permeability predictions were achieved for closely related analog series of compounds.
Quantitative structure property relationship (QSPR) has been recommended to predict human intestinal absorption without the need for actual compound synthesis. QSPR methods have been used to model physicochemical, chromatographic, and spectroscopic properties of compounds (Rubas et al., 1993; Artursson and Borchardt, 1997). Several computational methods have been described to predict the intestinal absorption parameters based on factors such as polar surface area, molecular surface area, dynamic surface area, etc. But most reports involve in sil-ico modeling studies performed on compounds closely related in structure thus making the model ineffective when applied to a wider structurally diverse data set. It is desirable to have predictive QSPR models that covers a diverse set of compounds with respect to their properties (e.g., physicochemical and pharmacological) as well as chemical structures.
Stenberg et al. (1999) compared the utility of three different predictive models for intestinal absorption. They demonstrated that molecular surface descriptors and descriptors derived from quantum mechanics were much more useful than the simple "rule of five" as the predictor of intestinal absorption. These in sil-ico methods have a great potential as virtual screens in testing permeability of test drugs. The utility of polar molecular surface area (PSAd) as a predictor of intestinal absorption was demonstrated by Clark (1999). Similarly, the evaluation of the dynamic PSAd as a predictor of drug absorption was performed by Palm et al. (1998). PSAd of the compounds was calculated from all low energy conformations identified in molecular mechanics calculations in vacuum and in simulated chloroform and water environment. PSAd was determined to be a better predictor compared to the octanol-water partition coefficient or the experimentally obtained immobilized liposome chromatography retention time.
Wessel et al. (1998) proposed an in silico method that used the molecular structural descriptors to predict the human intestinal absorption of drugs. Topological descriptors (based on 2D information of the compound), electronic descriptors (partial atomic charge and dipole moment), geometric descriptors (surface area, volume, etc.), and hybrid descriptors (combination of molecular surface area and partial atomic charge) were used for analysis. Based on the root-mean-square errors seen in the training set as well as the study set, it was demonstrated that there was a correlation between structure and intestinal absorption. The data set in the study included 86 structurally diverse compounds and the results confirm the potential of applying QSPR methods for estimating absorption. With further improvements in the choice of the descriptors, this in silico method can be used as a potential virtual screen in drug discovery processes. To date, one major critical impediment to a successful in silico modeling is the lack of a sufficiently large data base with reliable information. Also, the in silico methods, even at their best are not as reliable as real experimental data for predicting the permeability and absorption characteristics of compounds. However, in spite of the limitations, there are commercially available software packages (QMPRplus and Oraspotter) that incorporate the in silico methodology for predicting in vivo human absorption numbers.
One draw-back of these modeling efforts is that most published work is based on small sets of permeability data (< 100 compounds) that are collected from literature sources. Since the validity of any model depends on the data set used, a large and self-consistent data set is required for the development of global models with universal applicability. The quality and rigidity of the data set is also as important as the quantity of data. It is desirable to utilize the highest quality data to develop the models by using data generated from the same lab, using the same protocols and having appropriate controls. These in silico models, when carefully developed and rigorously validated, have the potential to be used in early screening set or library design. They can also play a role in lead optimization and preclinical candidate selection.
Was this article helpful?