Effects of Sample Size and Dimensionality on the Performance of Four Algorithms for Inference of Association Networks in Metabonomics

M. Suarez Diez*, E. Saccenti

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

18 Citations (Scopus)

Abstract

We investigated the effect of sample size and dimensionality on the performance of four algorithms (ARACNE, CLR, CORR, and PCLRC) when they are used for the inference of metabolite association networks. We report that as many as 100-400 samples may be necessary to obtain stable network estimations, depending on the algorithm and the number of measured metabolites. The CLR and PCLRC methods produce similar results, whereas network inference based on correlations provides sparse networks; we found ARACNE to be unsuitable for this application, being unable to recover the underlying metabolite association network. We recommend the PCLRC algorithm for the inference on metabolite association networks
Original languageEnglish
Pages (from-to)5119-5130
JournalJournal of Proteome Research
Volume14
Issue number12
DOIs
Publication statusPublished - 2015

Keywords

  • correlations
  • Low-molecular-weight metabolites
  • mutual information
  • network inference
  • network topology

Fingerprint

Dive into the research topics of 'Effects of Sample Size and Dimensionality on the Performance of Four Algorithms for Inference of Association Networks in Metabonomics'. Together they form a unique fingerprint.

Cite this