In 2002, our PreADME was submitted to The 14th European Symposium on Quantitative Structure-Activity Relationships (8th – 13th September 2002) at Bournemouth International Conference Centre, Bournemouth, UK.
This conference has been designed to review current practices used to identify new lead compounds and optimize their performance in industrial screens and as potential new products.
The PreADME approach: Web-based program for rapid prediction of physico-chemical, drug absorption and drug-like properties
Sung Kwang Lee, In Hye Lee, Hyun Ji Kim, Gue Su Chang, Jae Eun Chung, Kyoung Tai No
Computer Aided Molecular Design Research Center, Soong Sil University, Seoul 156-743, Korea
ADME, Absorption, logP, water solubility, molecular descriptor, neural network, QSAR
A significant bottleneck remains in the drug discovery procedure, in particular in the later stages of lead discovery, is analysis of the ADME and overt toxicity properties of drug candidates. Over 50% of the candidates failed due to ADME/Tox deficiencies during development. To avoid this failure at the development a set of in vitro ADME screens has been implemented in most pharmaceutical companies with the aim of discarding compounds in the discovery phase that are likely to fail further down the line. Even though the early stage in vitro ADME reduces the probability of the failure at the development stage, it is still time-consuming and resource-intensive. In this study, we describe a new web-based application called PreADME, which has been developed in response to a need for rapid prediction of drug-likeness and ADME data.
The PreADME resides entirely on a Web server, and can be accessed by browsers such as Netscape or Internet Explorer. PreADME consists of two main parts: a program called TOPOMOL, for calculation of important descriptors in characterizing ADME properties, and a neural network program, for the construction of drug absorption prediction system. The performance of the PreADME is as following:
- Molecular Descriptor Calculation:The ADME/Tox properties are closely related to physico-chemical descriptors such as lipophilicity (logP), molecular weight, polar surface area, and water solubility. The TOPOMOL program calculates more than 125 molecular descriptors including constitutional, topological, physico-chemical, and geometrical descriptors for ADME prediction from 2D chemical structure. TOPOMOL reads MDL mol or sd files and provides a rapid means to calculate 125 descriptors of 1,000,000 compounds in less than 1 hour using 1.6GHz PC.
- Drug-likeness Prediction:The most well known rule relating the chemical structures to their biological activities is Lipinski’s rule . Another well known rule is the Lead-like rule , i.e. starting from a qualitative survey based upon 18 lead and drug pairs of structure. Also, it is possible to provide any drug-like methods such as work of Ghose et al.  and Oprea et al. , or advanced customization option.
- ADME Prediction:In this study, the training set of structurally different compounds and their experimental permeability data were collected from the literature, in vitro Caco2-cell  and MDCK cell  assay. In addition, BBB (blood brain barrier) penetration, and HIA (human intestinal absorption) data were taken from ref [11-12], and [13-15], respectively. In drug absorption prediction, genetic functional approximation is used to select representative descriptors for absorption prediction, followed by back-propagation of error neural network to develop successful nonlinear model.
LogP and logS prediction
In this study, we tried to predict physico-chemical properties such as logP (octanol-water partition coefficient) and water solubility using a new atom-additive method (ASlogP and ASlogS) and to classify atoms by their hybridization states, their neighboring atom and some intramolecular interaction. The set of 6,943 experimental logP values and that of 1,403 experimental logS(water solubility) was fit with 215 adjustable atom types by linear least squares with r2=0.9276 and 0.9100, respectively. Compared to other methods (ClogP, ScilogP, AlogP98, SlogP and other papers), ours gives better predictions in logP and comparable prediction in logS.
Neural network analysis for absorption prediction (Caco-2, MDCK, BBB, HIA)
Currently, in vitro methods such as Caco-2 and MDCK monolayers or in vivo methods of HIA are widely used to estimate oral absorption. Similarly, there is a continuing interesting in developing a relevant in vitro screening for the penetration of BBB.
The following strategy was applied to obtain models that can be used to predict absorption data (Caco-2, MDCK, BBB, HIA): 1) calculate the molecular descriptors for the training and test set of compounds; 2) introduce genetic algorithm to find good descriptors set for training set compounds; 3) test predictability with test set compounds and determine the descriptors set for neural network training; 4) evaluate the performance of trained artificial neural network; 5) validate by blind testing.
Numerous neural network architectures were tested. The architecture with the lowest root-mean square error of test sets was deemed to be the optimal architecture. The optimum architecture of each prediction system, the training conditions for each neural network model are summarized in table 1.
Table 1. ANN Prediction Results of Different Absorption Data by the PreADME
|Model||System||¥ç||¥ì||Training set||Cross-Validation set|
We have shown that web-based program (PreADME) enables the calculation of molecular descriptors and prediction of drug-likeness and drug absorption data(Caco-2, MDCK, BBB, HIA) using 2D molecular structures, and without the need for special computational chemistry knowledge. PreADME is being widely used in our group, other institutes and companies (free of charge), and is starting to be adopted by medicinal chemist for design of chemical library used HTS and combinatorial chemistry.
- Kennedy,T. , Drug Discovery Today 1997, 2, 436-444.
- Ghose, A.K.; Viswanadhan,V.N.; Wendoloski, J.J. J. Phys. Chem. A 1998, 102, 3762-3772.
- Ertl, P.; Rohde, B.; Selzer, P. J. Med. Chem. 2000, 43, 3714-3717.
- Viswanadhan,V.N.;Ghose,A.K.;Singh,C.;Wendoloski,J.J. J. Chem. Inf. Comput. Sci. 1999, 39, 405-412.
- Lipinski,C.A.; Lombardo,F.; Dominy, B.W.; Feeney, P.J. Adv. Drug Deliv. Rev. 1997, 23, 3-25.
- Teague, S.J.; Davis, A.M.; Leeson P.D. ; Oprea, T. Angew. Chem. Int. Ed. 1999, 38, 3743.-3748
- Ghose, A.K.; Viswanadhan, V.N. and Wendoloski, J.J. J. Comb. Chem. 1999, 1, 55-68.
- Oprea, T.I. J. Comp-Aided Mol. Des. 2000, 14, 251-263.
- Kulkarni, A.; Han, Y. and Hopfinger, A. J. J. Chem. Inf. Comput. Sci,. 2002, 42, 331-342.
- Irvine, J. D. et al. J. Pharm. Sci. 1999, 88, 28-33.
- Rose, K. and Hall, L. H. J. Chem. Inf. Comput. Sci,. 2002, 42, 651-666.
- Kelder, J.; Grrotenhuis, P.D.J.; Bayada, D.M.; Delbressine, L.P.C.; Ploemen,J.-P. Pharm. Res. 1999, 16(10), 1514-1519.
- Zhao, Y. H.; Le, J.; Abraham, M. H. et al. J. Pharm. Sci. 2001, 90(6), 749-784.
- Wessel, M. D.; Jurs, P. C.; Tolan, J.W.; Muskal, S.M. J. Chem. Inf. Comput. Sci. 1998, 38, 726-735.
- Raevsky, O.A.; Schaper, K.-J.; Artursson, P.; McFarland, J.W. Quant. Struct. -Act. Relat., 2002, 20, 402-413.