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

Analysis of linear and nonlinear QSAR data using neural networks

D T Manallack1, D D Ellis, D J Livingstone

  • 1SmithKline Beecham Pharmaceuticals, Welwyn, Herts, U.K.

Journal of Medicinal Chemistry
|October 28, 1994
PubMed
Summary

Feed forward back propagation neural networks show promise for regression tasks but struggle with predictive accuracy in QSAR studies. These networks are prone to chance effects and lack interpretability, limiting their practical application.

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

IRAK3 modulates downstream innate immune signalling through its guanylate cyclase activity.

Scientific reports·2019
Same author

The acid/base profile of agrochemicals.

SAR and QSAR in environmental research·2017
Same author

Multiple Bud Formation by Cultured Embryos of Pinus ponderosa.

Journal of plant physiology·2012
Same author

Use of Kv1.3 blockers for inflammatory skin conditions.

Current medicinal chemistry·2010
Same author

Computational techniques for the prediction of toxicity.

Toxicology in vitro : an international journal published in association with BIBRA·2010
Same author

Synthesis and preliminary screening of novel tryptamines as 5-HT4 receptor ligands.

Current medicinal chemistry·2010

Area of Science:

  • Computational Chemistry
  • Cheminformatics
  • Machine Learning

Background:

  • Artificial neural networks (ANNs) are increasingly used in scientific research.
  • Feed forward back propagation networks offer an alternative to traditional statistical methods like multiple linear regression.

Purpose of the Study:

  • To evaluate the performance of feed forward back propagation neural networks for quantitative structure-activity relationship (QSAR) modeling.
  • To assess the predictive ability and interpretability of these ANNs compared to established methods.

Main Methods:

  • Utilized artificial structured datasets and real literature data for network training and testing.
  • Employed leave-one-out cross-validation and training/test set protocols to assess predictive performance.

Related Experiment Videos

Main Results:

  • Neural networks demonstrated a strong ability to fit the training data.
  • However, poor predictive performance was observed for the QSAR data.
  • Identified susceptibility to chance effects and difficulties in interpreting the developed relationships.

Conclusions:

  • Feed forward back propagation neural networks, while capable of data fitting, present significant limitations for QSAR prediction.
  • Their susceptibility to chance and lack of interpretability hinder reliable application.
  • Further research into alternative ANN architectures and applications is warranted.