A comparative evaluation of data-merging and meta-analysis methods for reconstructing gene-gene interactions

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Analyzing multiple gene expression datasets aids in reconstructing gene-gene interaction networks.
• Systematic variations across studies pose challenges for accurate network inference.
• Meta-analysis and data-merging are common strategies to address these biases.

Purpose of the Study:

• To compare the effectiveness of meta-analysis and data-merging approaches for gene expression data integration.
• To evaluate methods for reconstructing gene-gene interaction networks from heterogeneous microarray data.
• To assess the impact of systematic variations on network inference.

Main Methods:

• Comparative evaluation of meta-analysis and data-merging techniques.
• Analysis of synthetic and real gene expression data, including E. coli and Yeast microarray compendia.
• Case-study on reconstructing the Ikaros transcription factor regulatory network in human PBMCs.
• Utilized correlation statistics and p-values for ranking candidate interactions.

Main Results:

• Meta-analysis and data-merging methods demonstrated comparable performance on both synthetic and real data.
• Both integration methods significantly outperformed naive data merging and single-dataset analysis.
• Correlation statistics were more effective than p-values for identifying true interactions.
• Findings from the PBMC case-study confirmed generalizability across network reconstruction algorithms.

Conclusions:

• Ignoring systematic variations in heterogeneous gene expression studies leads to unreliable network reconstruction.
• Meta-analysis and data-merging methods are equally effective in mitigating biases.
• Researchers can choose the method that best fits their specific application for accurate network analysis.