DExMA

DExMA performs meta-analysis of gene expression studies to integrate heterogeneous transcriptomic datasets and identify reproducible biomarkers and cross-study expression signatures.

Key Features:

• R/Bioconductor implementation: Implemented as an R/Bioconductor package for transcriptomic meta-analysis.
• Automated data retrieval: Downloads expression matrices directly from GEO accession codes.
• Missing-gene handling: Provides two strategies for missing genes: sampkeKNN, a k-nearest neighbours–based imputation using samples from other studies, and user-controlled filtering based on the minimum proportion of datasets in which a gene must be observed.
• Meta-analysis models: Implements multiple meta-analytic methodologies and models suitable for transcriptomic data integration.
• Quality control and heterogeneity assessment: Performs quality control and assesses heterogeneity across combined datasets.
• Statistical integration and visualization: Performs statistical integration of results and provides visualization of meta-analysis outputs.
• Cross-platform support: Addresses platform-specific probe-set differences when combining heterogeneous datasets.

Scientific Applications:

• Increasing statistical power: Combines independent transcriptomic studies to increase power for detecting differential expression.
• Reproducible biomarker identification: Identifies reproducible biomarkers across independent cohorts.
• Cross-disease comparisons: Enables cross-disease comparisons to uncover shared or inverse expression signatures.
• Case–control meta-analysis: Applicable to case–control study designs for differential expression meta-analysis.
• Concordant/discordant pattern detection: Detects concordant or discordant gene-expression patterns across independent cohorts.

Methodology:

Implemented in R/Bioconductor with automated download of expression matrices from GEO accession codes, quality control and heterogeneity assessment, multiple meta-analysis models, missing-gene handling via sampkeKNN k-nearest neighbours–based imputation and user-controlled minimum-dataset filtering, and statistical integration with visualization of results.