|
Internet Electronic Journal of Molecular Design - IEJMD, ISSN 1538-6414, CODEN IEJMAT
| ABSTRACT - Internet Electron. J. Mol. Des. February 2005, Volume 4, Number 2, 124-150 |
Prediction of Intestinal Epithelial Transport of Drug in (Caco-2)
Cell Culture from Molecular Structure using in silico
Approaches During Early Drug Discovery
Yovani Marrero Ponce, Miguel A. Cabrera Pérez, Vicente Romero Zaldivar, Marival Bermejo Sanz, Dany Siverio Mota, and Francisco Torrens
Internet Electron. J. Mol. Des. 2005, 4, 124-150
|
Abstract:
The high interest in the prediction of the intestinal absorption for
new chemical entities is generated by the increasing rate in the
synthesis of compounds by combinatorial chemistry and the
extensive cost of the traditional evaluation methods. Novel
molecular descriptors have been applied to estimate the intestinal
epithelial transport of drug in Caco-2 cell culture. Total and local
(atom and atom-type) quadratic indices used in this study were
calculated by TOMOCOMD-CARDD software. Linear
Discriminant Analysis (LDA) was used to obtain a quantitative
model that discriminates the high absorption compounds (P ≥
8×10-6 cm/s) from those with moderate-poor absorption (P <
8×10-6 cm/s). A data set of 134 diverse structure drugs and two
series of drugs-like compounds (12 compounds) were used as
training and test set, respectively. In addition, Multiple Linear
Regression (MLR) has been carried out to derive QSPerR
models. All statistical analyses were performed with the
STATISTICA software package. The obtained LDA model
classified correctly 81.13% of compounds with moderate-poor
absorption properties and the 96.30% of compounds with high
absorption, showing a global good classification of 90.30% in the
training set. The model showed a high Matthews' correlation
coefficient (MCC = 0.80). Internal and external validation
processes to demonstrate the robustness and predictive power of
the obtained model were carried out. In this sense, the model
classified correctly 87.31% (MCC = 0.73) in the leave-one-out
cross-validation procedure. The discriminant model was also
assessed by a 10-fold full cross-validation (removing
approximately 13 compounds in each cycle, 85.82% of good
classification), yielding a MCC of 0.70. Also this model shown
an 87.5, 85.6, 84.7, 85.0, 85.3, 83.5, 84.1, 86.2, 85.9 and 85.9%
of global good classification when n varied from 2 to 11 in the
leave-n-out cross validation procedure. The model was stabilized
around 85.9% when n was > 9. In addition, a data set of 7 HIV
protease inhibitors (4 linear peptidomimetic and 3 new cyclic
urea) and 5 new 6-fluoroquinolones derivatives was used as
external test set. The LDA-QSPerR model achieved a MCC of
0.71 (83.33% correct prediction) in this study. This approach
permits us to obtain a good explanation of the experiment based
on the molecular structural features, evidencing the main role of
H-bonding and size properties in permeability process. Finally,
the model developed was used in the virtual screening of 241
drugs with the percentage of human intestinal absorption (Abs
%) values reported. A relationship between the predicted
permeability coefficients in Caco-2 and the Abs % (145
compounds with good data quality) was established, with a
percentage of good relation greater than 82 %. A comparison
with results derived from other three theoretical studies shown a
quite satisfactory behavior of the present method. All these
results shown that total and local (atom and atom-type) quadratic
indices can successfully predict intestinal permeability and
suggest that the proposed methodology will be a good tool for
studying the oral absorption of drug candidates during the drug
development process.
|