Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

A comparative study on feature selection methods for drug discovery.

Ying Liu1

  • 1Georgia Institute of Technology, College of Computing, Atlanta, Georgia 30322, USA. yingliu@cc.gatech.edu

Journal of Chemical Information and Computer Sciences
|September 28, 2004
PubMed
Summary

Feature selection in drug discovery significantly improves Naïve Bayesian classifiers, especially using information gain. Support Vector Machines (SVM) performed best with all features, though information gain minimally impacted SVM performance.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

[Preparation and identification of monoclonal antibody against Homo sapiens hemoglobin alpha 2 (HBA2)].

Zhongguo shi yan xue ye xue za zhi·2007
Same author

[A case-control study on the relations between insulin resistance and essential hypertension in YI NATIONALITY living in Liangshan, Sichuang Province].

Zhonghua yu fang yi xue za zhi [Chinese journal of preventive medicine]·2007
Same author

Growth enhancement effect of BzATP on primary cultured astrocytes from rat brain.

Neuroscience bulletin·2007
Same author

Total and high molecular weight but not trimeric or hexameric forms of adiponectin correlate with markers of the metabolic syndrome and liver injury in Thai subjects.

The Journal of clinical endocrinology and metabolism·2007
Same author

The novel sandwich-type heteropolyoxotungstates [M2Bi2(beta-B-MW9O34)2]14- (M = Co(II), Zn(II)): beta-type dimeric heteropolyanions with a transition metal as the central heteroatom and Bi(III) and M as linking atoms.

Dalton transactions (Cambridge, England : 2003)·2007
Same author

Threshold-style processing of Chinese characters for adult second-language learners.

Memory & cognition·2007

Area of Science:

  • Computational chemistry
  • Machine learning in drug discovery

Background:

  • Feature selection is a crucial preprocessing step in machine learning.
  • Reducing irrelevant and redundant data can enhance algorithm performance.
  • This study focuses on aggressive dimensionality reduction for feature selection in drug discovery.

Purpose of the Study:

  • To comparatively evaluate five feature selection methods in the context of drug discovery.
  • To assess the impact of feature selection on Naïve Bayesian and Support Vector Machine (SVM) classifiers.
  • To identify the most effective feature selection techniques for chemical compound classification.

Main Methods:

  • Evaluated five feature selection methods: information gain, mutual information, chi2-test, odds ratio, and GSS coefficient.

Related Experiment Videos

  • Utilized Naïve Bayesian and Support Vector Machine (SVM) algorithms for classifying chemical compounds.
  • Performed aggressive dimensionality reduction, removing up to 99% of features.

    • Main Results:

    • Naïve Bayesian classifiers showed significant performance improvement with feature selection, particularly using information gain and chi2-test.
    • Information gain with Naïve Bayesian achieved improved classification accuracy (sensitivity) even after removing 96% of features.
    • Support Vector Machines (SVM) performed optimally with all features; however, they were less sensitive to feature reduction when using information gain, maintaining high specificity.
    • Mutual information performed poorly due to bias towards rare features and sensitivity to estimation errors.

    Conclusions:

    • Information gain and chi2-test are highly effective feature selection methods for drug discovery classification tasks, especially with Naïve Bayesian classifiers.
    • Aggressive dimensionality reduction using feature selection can lead to improved classification accuracy and efficiency without substantial performance loss for certain algorithms.
    • The choice of feature selection method and classifier significantly impacts performance in drug discovery applications.