Related Experiment Video
Updated: May 22, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
Improving biomarker list stability by integration of biological knowledge in the learning process
Tiziana Sanavia1, Fabio Aiolli, Giovanni Da San Martino
1Department of Information Engineering, University of Padova, via G, Gradenigo 6/B, 35131 Padova, Italy.
BMC Bioinformatics
|April 28, 2012
Summary
Integrating biological information enhances the stability of molecular biomarker discovery lists. Gene Ontology annotations and protein-protein interactions improve list consistency without sacrificing prediction accuracy for disease diagnosis.
Area of Science:
- Bioinformatics
- Genomics
- Systems Biology
Background:
- Identifying robust molecular biomarkers is crucial for early disease diagnosis and treatment.
- Current microarray data analysis methods for biomarker discovery often yield inconsistent results.
- Complex diseases involve multiple genes and pathways, complicating biomarker identification.
Purpose of the Study:
- To assess the impact of integrating different types of biological information on biomarker discovery list stability.
- To compare the effectiveness of functional annotations, protein-protein interactions, and gene expression correlations.
Main Methods:
- Developed gene similarity matrices based on Gene Ontology (GO) annotations (Biological Process, Molecular Function).
- Applied geodesic distance on protein-protein interaction (PPI) networks.
- Transformed expression data to reflect feature similarity.
Main Results:
- Semantic similarity matrices from GO annotations and PPI network analysis significantly improved biomarker list stability.
- These methods maintained high prediction accuracy.
- Gene Ontology and protein-protein interaction data integration proved most effective.
Conclusions:
- Integrating biological knowledge, particularly through GO and PPI networks, enhances biomarker list stability.
- Features (genes) with strong correlations, like those in PPI networks, may share similar relevance.
- Findings provide a basis for further research into combining similarity matrices for more stable biomarker lists.
