Interpretable prediction of non-genotoxic hepatocarcinogenic chemicals

Chun Wei Tung, Jhao Liang Jheng

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

The assessment of non-genotoxic hepatocarcinogenicity of chemicals relies on time-consuming rodent bioassays. The development of alternative methods for non-genotoxic hepatocarcinogenicity could help the identification of potential hepatocarcinogenic chemicals. This study evaluated four types of features for the interpretable prediction of non-genotoxic hepatocarcinogenic chemicals including chemical-chemical interactions (CCI), chemical-protein interactions (CPI), chemical descriptors (QSAR) and gene expression profiles (TGx). Based on the results of decision tree classifiers, the CPI-based features perform best with independent test accuracies of 90% and 86% for interaction scores from combined scores and databases, respectively. Informative features were identified and analyzed to give insights into the non-genotoxic hepatocarcinogenicity of chemicals. The difference between CPI scores and gene expression profiles for the identified important proteins shows that CPI could play more important roles in non-genotoxic hepatocarcinogenicity.

Original languageEnglish
Pages (from-to)68-74
Number of pages7
JournalNeurocomputing
Volume145
DOIs
Publication statusPublished - Dec 5 2014
Externally publishedYes

    Fingerprint

Keywords

  • Chemical-chemical interaction
  • Chemical-protein interaction
  • Decision tree
  • Non-genotoxic hepatocarcinogenicity
  • Quantitative structure-activity relationship
  • Toxicogenomics

ASJC Scopus subject areas

  • Computer Science Applications
  • Cognitive Neuroscience
  • Artificial Intelligence

Cite this