targetedRetrieval

targetedRetrieval retrieves gene expression experiments by measuring similarity based on inferred regulatory relationships among specified genes to contextualize new experiments within public gene expression repositories.

Key Features:

• Targeted focus and interpretability: Centers analysis on regulatory relationships for analyst-specified genes rather than global gene-set similarity or experimental annotation, yielding more interpretable retrieval results.
• Regulatory model-based similarity: Learns a regulatory model from gene expression repositories that captures inferred regulatory links between genes and uses that model to compare experiments.
• Fisher kernel similarity: Employs the Fisher kernel as a rigorous similarity measure derived from the learned generative regulatory model.
• Improved retrieval performance: Demonstrates significant enhancement in retrieving related experiments compared to standard methods across human and plant microarray datasets.
• Biological interpretation via link activity: Highlights changes in regulatory link activity patterns to facilitate interpretation of biological conditions, exemplified by identification of key relationships in the osmotic stress network of Arabidopsis thaliana.

Scientific Applications:

• Contextualizing new experiments: Positions new gene expression experiments within the landscape of public expression repositories to identify related studies.
• Regulatory interaction analysis: Focuses on pre-specified genes to investigate and retrieve experiments relevant to specific regulatory interactions.
• Cross-organism microarray retrieval: Applicable to both human and plant microarray datasets for comparative retrieval of related experiments.

Methodology:

Learn a regulatory model from gene expression data repositories, use the model to derive a Fisher kernel-based similarity metric from inferred regulatory links, and apply that metric to retrieve similar experiments.