ExpressionBlast

ExpressionBlast performs large-scale similarity search and comparative analysis of gene expression profiles across heterogeneous public repositories to identify related perturbations, pathways, phenotypes, and disease states.

Key Features:

• Metadata extraction and classification: Automates metadata extraction, replicate identification, and treatment–control classification using text mining and heuristic matching.
• Probe harmonization: Harmonizes probe identifiers across experiments and platforms.
• Expression scaling and normalization: Converts expression values to consistent log-ratio scales and normalizes distributions.
• Large GEO index: Indexes >900,000 samples from >40,000 Gene Expression Omnibus (GEO) series.
• Similarity scoring: Computes similarity scores using distance metrics such as Euclidean distance, correlation, and anti-correlation.
• Cross-species orthology: Incorporates orthology mappings via InParanoid and resolves ambiguous mappings by closest expression similarity.
• Statistical augmentation and enrichment: Augments matches with statistical significance estimates, GO term enrichment, and key shared text-mined keywords, and links to supporting publication abstracts.
• Visualization: Visualizes results as heatmaps.

Scientific Applications:

• Perturbation and phenotype discovery: Identifies related perturbations, pathways, phenotypes, and disease states by matching expression signatures across datasets.
• Cross-platform and cross-species comparison: Enables comparative analysis across platforms and species using probe harmonization and InParanoid orthology mappings.
• Functional genomics and systems biology: Supports hypothesis generation and interpretation in functional genomics and systems biology by mining public expression archives.
• Meta-analysis and literature linking: Facilitates meta-analysis of GEO datasets and links expression signatures to supporting publication abstracts and enriched GO terms.

Methodology:

Metadata extraction, replicate identification, and treatment–control classification via text mining and heuristic matching; harmonization of probe identifiers; conversion of expression values to log-ratio scales and distribution normalization; indexing of >900,000 GEO samples; similarity scoring using Euclidean distance, correlation, and anti-correlation; orthology mapping via InParanoid with ambiguous mappings resolved by closest expression similarity; computation of statistical significance estimates, GO enrichment, and extraction of shared text-mined keywords with links to publication abstracts.