Discovering Pair-wise Synergies in Microarray Data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate gene selection is crucial for cancer diagnosis and drug target identification.
• Existing methods often overlook complex gene interactions, limiting their effectiveness.
• Current information-theoretic approaches for gene synergy detection rely on assumptions about data characteristics.

Purpose of the Study:

• To develop and validate an extension of the Maximal Information Coefficient (MIC) to estimate MIC(X1; X2; Y) for detecting pair-wise gene synergy.
• To assess the generality and effectiveness of the proposed MIC(X1; X2; Y) measure in identifying synergistic genes.
• To demonstrate the utility of MIC(X1; X2; Y) in cancer diagnosis and drug discovery by analyzing simulation and microarray data.

Main Methods:

• Developed an approximation algorithm for estimating MIC(X1; X2; Y) for a discrete variable Y.
• Applied MIC(X1; X2; Y) to detect pair-wise synergy in simulated datasets and real cancer microarray data.
• Compared the performance of MIC(X1; X2; Y) against established feature selection methods like MIC(X; Y) and TSG.

Main Results:

• The proposed MIC(X1; X2; Y) measure exhibits generality in capturing a wide range of associations.
• MIC(X1; X2; Y) successfully identified synergistic genes that were undetectable by reference methods.
• Identified synergistic genes demonstrated the ability to distinguish between different phenotypes.
• The biological relevance of discovered synergistic genes was confirmed through Gene Ontology (GO) annotation and the OUgene database.

Conclusions:

• MIC(X1; X2; Y) provides a powerful and generalizable approach for discovering synergistic gene pairs.
• This method enhances the identification of informative genes for cancer diagnosis and potential drug targets.
• The findings highlight the importance of considering higher-order gene interactions for biological discovery.