Premium
Integrated One‐Against‐One Classifiers as Tools for Virtual Screening of Compound Databases: A Case Study with CNS Inhibitors
Author(s) -
JalaliHeravi Mehdi,
ManiVarnosfaderani Ahmad,
Valadkhani Abolfazl
Publication year - 2013
Publication title -
molecular informatics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.481
H-Index - 68
eISSN - 1868-1751
pISSN - 1868-1743
DOI - 10.1002/minf.201200126
Subject(s) - pubchem , virtual screening , cheminformatics , support vector machine , artificial intelligence , linear discriminant analysis , computer science , chemical space , machine learning , quadratic classifier , data mining , receiver operating characteristic , pattern recognition (psychology) , drugbank , multiclass classification , drug discovery , bioinformatics , computational biology , biology , drug , pharmacology
Abstract A total of 21 833 inhibitors of the central nervous system (CNS) were collected from Binding‐database and analyzed using discriminant analysis (DA) techniques. A combination of genetic algorithm and quadratic discriminant analysis (GA‐QDA) was proposed as a tool for the classification of molecules based on their therapeutic targets and activities. The results indicated that the one‐against‐one (OAO) QDA classifiers correctly separate the molecules based on their therapeutic targets and are comparable with support vector machines. These classifiers help in charting the chemical space of the CNS inhibitors and finding specific subspaces occupied by particular classes of molecules. As a next step, the classification models were used as virtual filters for screening of random subsets of PUBCHEM and ZINC databases. The calculated enrichment factors together with the area under curve values of receiver operating characteristic curves showed that these classifiers are good candidates to speed up the early stages of drug discovery projects. The “relative distances” of the center of active classes of biosimilar molecules calculated by OAO classifiers were used as indices for sorting the compound databases. The results revealed that, the multiclass classification models in this work circumvent the definition inactive sets for virtual screening and are useful for compound retrieval analysis in Chemoinformatics.