|
Internet Electronic Journal of Molecular Design - IEJMD, ISSN 1538-6414, CODEN IEJMAT
| ABSTRACT - Internet Electron. J. Mol. Des. November 2006, Volume 5, Number 11, 542-554 |
Artificial Immune Systems in Drug Design: Recognition of P-Glycoprotein Substrates
with AIRS (Artificial Immune Recognition System)
Ovidiu Ivanciuc
Internet Electron. J. Mol. Des. 2006, 5, 542-554
|
Abstract:
Artificial immune systems (AIS) represent a new class of machine
learning procedures that simulate several mechanisms and functions of
the biological immune system, such as pattern recognition, learning,
memory, and optimization. In this paper we present the first
application of the artificial immune recognition system (AIRS) to the
recognition of the substrates of the multidrug resistance (MDR) ATP-
binding cassette (ABC) transporter permeability glycoprotein (P-glycoprotein,
P-gp). We evaluated the AIRS algorithm for a dataset of
201 chemicals, consisting of 116 P-gp substrates and 85 P-gp
nonsubstrates. The classifiers were computed from 159 structural
descriptors from five classes, namely constitutional descriptors,
topological indices, electrotopological state indices, quantum
descriptors, and geometrical indices. The AIRS algorithm is controlled
by eight user defined parameters: affinity threshold scalar, clonal rate,
hypermutation rate, number of nearest neighbors, initial memory cell
pool size, number of instances to compute the affinity threshold,
stimulation threshold, and total resources. The AIRS sensitivity to
these parameters was investigated with leave-20%-out (five-fold)
cross-validation predictions performed over a wide range of values for
the eight AIRS parameters. The AIRS algorithm (best predictions:
selectivity 0.793, specificity 0.577, accuracy 0.702, and Matthews
correlation coefficient 0.380) was compared with 13 well-established
machine learning algorithms. The AIRS predictions are better than
those of five of these algorithms (alternating decision tree, Bayesian
network, logistic regression with ridge estimator, random tree, and fast
decision tree learner), showing that P-gp substrates may be
successfully recognized with AIRS. In conclusion, classifiers based on
artificial immune systems are valuable tools for structure-activity
relationships (SAR), quantitative structure-activity relationships
(QSAR), drug design, and virtual screening of chemical libraries.
|