Enhanced replacement method integration with genetic algorithms populations in QSAR and QSPR theories

MERCADER, A. G.; DUCHOWICZ, P. R.

CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS

ELSEVIER SCIENCE BV

Lugar: Amsterdam; Año: 2015 vol. 149 p. 117 - 122

0169-7439

The selection of an optimal set ofmolecular descriptors fromamuch larger collection of such regression variablesis a vital step in the elaboration ofmost QSAR and QSPRmodels. The aimof thiswork is to continue advancing thisimportant selection process by combining the enhanced replacement method (ERM) and the well-knowngenetic algorithms (GA). These approaches had previously proven to yield near-optimal results with a muchsmaller number of linear regressions than a full search. The newly proposed algorithms were tested on fourdifferent experimental datasets, formed by collections of 116, 200, 78, and 100 experimental records fromdifferent compounds and 1268, 1338, 1187, and 1306 molecular descriptors, respectively. The comparisonsshowed that the new alternative ERMp (combination of ERM with a GA population) further improves ERM, ithas previously been shown that the latter is superior to GA for the selection of an optimal set of molecular descriptorsfrom a much greater pool.