Article Preview
TopIntroduction
In drug development, a large amount of possible drug candidate molecules, called “lead compounds”, may be predicted through drug design and computational modeling. However, about 95% of lead compounds have failed in the development stages, and 50% of these failures were shown to be due to unfavorable absorption, distribution, metabolism, and excretion (ADME) properties (Beresford et al., 2002). In the development of drugs intended for oral use, good drug absorption and appropriate drug delivery are very important (Hou & Xu, 2004).
Inadequate bioavailability is one of the main reasons that cause many promising drug candidates failed in clinical trials. Bioavailability represents the percentage of an oral dose which is able to produce a pharmacological activity, in other words, the fraction of the oral dose that reaches the arterial blood in an active form. Oral bioavailability is related to several factors, such as gastrointestinal transition and absorption, intestinal membrane permeation, and intestinal/hepatic first-pass metabolism. Moreover, during absorption, many researchers have suggested that gut wall Cytochrome P450 3A4 and P-glycoprotein, the multidrug transporter, act in a concerted manner to control the absorption of their substrates (Van Asperen et al., 1998; Lampen et al., 1998; Hall et al., 1999). One possible method to maximize oral absorption would be to design a molecule that acts as a substrate of P-glycoprotein and CYP3A4 (Van de Waterbeemd, 2001).
Some researchers have summarized general molecular properties of drug molecules that may lead to good drug absorbency. Veber et al. (2002) reported studies on rat bioavailability data for 1100 drug candidates. It was found that drug molecules having fewer than 10 rotatable bonds and less than 140 Å2 PSA (Polar Surface Area) (or a hydrogen bond count less than 12) usually showed more than 20% rat oral bioavailability. Lu et al. (2004) investigated the relationship between number of rotatable bonds and PSA for rat oral bioavailability of 434 molecules. Compared to Veber’s work (Veber et al., 2002), Lu et al. reported that the prediction results were dependent on the calculation methods.
There are also quantitative studies in order to predict bioavailability. Hirono et al. (1994) reported a quantitative structure-bioavailability model for 188 noncongeneric organic drugs; the drugs were separated into three groups: nonaromatics, aromatics, and heteroaromatics. Based on each group’s chemical and physical properties, a quantitative model was developed. Andrews et al. (2000) reported a QSAR (Quantitative Structure and Activity Relationship) model based on 591 compounds and 85 descriptors. The model achieved a good correlation (r2=0.71), but over-fitting problem may exist from the cross-validation result (q2=0.58).
Yoshida et al. (2000) published a classification model for human oral bioavailability. This model can get a correct rate of 60% for the test group. Turner et al. (2003) reported a QSAR model for a dataset of 169 compounds using stepwise regression method. The regression model were built based on a training set including 159 compounds and validated by a test set including 10 compounds. Although the correlation coefficient of 0.72 was obtained for the test set, due to the small size of the test set, the validation was not statistically reliable. Turner et al. (2004) developed another artificial neural network (ANN) model for oral bioavailability. The correlation coefficients for training set and test set of the model were 0.736 and 0.897, respectively. Wang et al. (2006) reported a correlation model for the predictions of oral bioavailability. The models were built based on the counts of molecular fragments descriptors using a genetic algorithm (GA) and consensus score was employed for a molecule which was simply defined as the average of the values predicted by the best 20 GA-QSAR models. However, the prediction capability of this model is also questionable due to the low correlation coefficient value of cross-validation.
Hou et al. (2007) conjectured that there are no simple rules based on molecular descriptors that can be used to predict human oral bioavailability accurately and reliably by analyzing a diverse dataset of 773 compounds with experimental human oral bioavailability values.