Substructure‐Based Support Vector Machine Classifiers for Prediction of Adverse Effects in Diverse Classes of Drugs.

Abstract

BHAVANI, S.; NAGARGADDE, A.; THAWANI, A.; SRIDHAR, V.; CHANDRA*, N.; J. Chem. Inf. Model. (J. Chem. Inf. Comput. Sci.) 46 (2006) 6, 2478-2486; Bioinf. Cent., Indian Inst. Sci., Bangalore 560 012, India; Eng.) -Lindner 07-215

Predicting chemical activities from structures by attributed molecular graph classification

“…Genetic Programming, an evolutionary algorithm-based methodology, has been used to model drug bio-availability [15], [16], [17] and has been developed to predict p450 inhibition [18], [19] coupled with ANNs. Support Vector Machine has been popularly used in drug virtual screening [4], [5], [6]. [7], [8] studied the "drug-likeness" problem and claimed that SVM predictions were more robust than those from neural networks and [9] employed SVM to predict chemists' intuitive assessments.…”

“…Furthermore, in recent years, kernel methods, in particular Support Vector Machine (SVM), have become increasingly popular and important for designing QSAR models due to its relatively high predictive strength [4], [5], [6], [7], [8], [9]. An important tool in using kernel methods is the notion of a kernel function, which can be thought as a special similarity measure between input data and can be guaranteed to be able to map onto a Hilbert space.…”

Machine learning algorithms for the prediction of hERG and CYP450 binding in drug development

“…A rank-by-vote consensus scoring approach was used to classify the data into three categories: hERG binders, nonbinders and intermediates. Bhavani et al [16] 2006 271 78 FSG [61] (1256 total) SVM Binary Acc: 93.6%…”

“…Bhavani et al have used a substructure-based approach to train an SVM model to distinguish between compounds that are either known to cause torsades de point (TdP+) or do not induce torsades de point (TdP-) [16] using a dataset reported by Yap et al [17]. The authors used the Frequent SubGraph (FSG) algorithm to identify the frequently occurring substructures based on the topological graphs of the compounds in the dataset.…”

Calculation and application of activity discriminants in lead optimization

“…The number of data points in each assay is number of compounds that have been successfully measured experimentally. The number of compounds in the test set is the number of compounds synthesized in 2006-2008, used [23][24][25][26], physical chemical properties [27], partial charge [28][29][30], topological polar surface area [31][32][33], molecular fingerprints [34][35][36][37][38][39][40], molecular connectivity indices [41][42][43][44] and E-state indices [44][45][46][47], CoMFA [48,49], GRIND [50][51][52], and Volsurf [53][54].…”