TL;DR
STARNET 2 is a new web-based tool that helps researchers discover gene regulatory networks by visualizing co-expression patterns from large microarray datasets across ten species. It addresses the challenge of interpreting complex gene expression data by providing interactive network graphs, known interaction data, and statistical enrichment analyses, making it easier to generate hypotheses about gene regulation.
What Problem Does This Paper Address?
Analyzing gene expression data from microarrays, while powerful, presents significant challenges. Biologists often end up with lists of differentially expressed genes but struggle to ascribe biological meaning to them. This involves inferring relationships between genes and understanding how they contribute to biological processes. Comparing data across different experiments, platforms, and even species further complicates this, requiring extensive data processing. Existing tools often focus on specific experimental questions or lack retrospective analysis capabilities. STARNET 2 aims to bridge this gap by providing a user-friendly platform for exploring existing microarray data to uncover potential gene regulatory relationships and generate new hypotheses without needing to perform costly new experiments.
Methods
STARNET 2 leverages pre-computed pairwise Pearson correlation coefficients derived from a large collection of microarray expression data. The core methodology involves:
- Data Collection and Preprocessing: Raw microarray data was retrieved from NCBI's Gene Expression Omnibus (GEO) for ten different species. Data was normalized using the JustRMALITE method.
- Correlation Network Construction: Pairwise Pearson correlation coefficients were computed for gene expression profiles within each species and platform. These correlations were stored in a MySQL database.
- Web-Based Query Interface: Users can input a gene of interest (by Entrez Gene ID or symbol) and select one or two data cohorts (e.g., "full" or "developmental" for an organism, or across species).
- Network Visualization: STARNET 2 generates interactive graphs of correlation networks centered on the gene of interest. Node colors represent network levels, and line colors indicate the strength of correlation.
- Enrichment Analysis: The tool performs an on-the-fly test for Gene Ontology (GO) term enrichment within the identified networks to highlight potentially functionally related genes.
- Known Interaction Integration: The system supplements correlation networks with known interactions involving genes and gene products, retrieved from NCBI databases.
- HEATSEEKER Module: This new module allows for direct visual comparison of two networks (from different cohorts or species) using false-color heat maps of correlation distances. It also provides statistical significance of differences in correlations.
Datasets
The study utilized data from the NCBI Gene Expression Omnibus (GEO). For each of the ten species, data was collected from between 148 (rice) and 3,763 (human) Affymetrix microarray samples. A total of 12,762 arrays were used, representing approximately 5% of the samples available in GEO at the time of the study (August 2008). Specific details on array platforms and selected experiments are available via provided links.
Key Insights and Ideas
- Leveraging Public Data: STARNET 2 effectively harnesses the vast amount of publicly available microarray data for retrospective analysis, enabling hypothesis generation without new experimental costs.
- Co-expression as a Proxy for Regulation: The tool's foundation on Pearson correlation coefficients assumes that genes with similar expression patterns are likely to be involved in related biological processes or regulatory pathways.
- Multi-Species Comparative Analysis: The expansion to ten species and the introduction of the HEATSEEKER module are crucial for comparative genomics, allowing researchers to explore conserved or divergent regulatory mechanisms across different organisms.
- Integrated Analysis: Combining correlation networks with known interaction databases and GO enrichment analysis provides a more comprehensive view, helping to validate and interpret potential regulatory relationships.
- User-Centric Discovery: The interactive and visual nature of STARNET 2 empowers researchers to explore data intuitively, allowing for flexible queries and network exploration tailored to their specific interests.
Key Results
STARNET 2 provides a user-friendly interface for exploring gene regulatory networks. For any given gene, it generates correlation networks based on user-selected data cohorts. These networks are supplemented with lists of genes and their correlation values, links to Entrez Gene entries, and GO term enrichment results. The HEATSEEKER module allows for visual comparison of networks, highlighting differences in co-expression patterns between cohorts or species. The tool is freely available and does not require user registration.
Limitations and Future Work
- Correlation vs. Causation: The paper acknowledges that correlation does not imply causation. While STARNET 2 identifies potential regulatory relationships, further experimental validation is necessary.
- Quantitative Validation: The authors suggest that future studies should quantitatively assess STARNET's ability to recapitulate known biological networks to understand the extent to which its identified edges can predict regulatory relationships.
- Differential Wiring: While STARNET 2 facilitates comparisons that can hint at differential wiring, the authors note that the HEATSEEKER module is a "first step" towards more careful and unbiased comparisons.
- Heuristics in Other Methods: The paper positions STARNET 2 as an attractive alternative to methods like Bayesian networks, which can be computationally expensive and often require time-series data. However, it implies that STARNET 2's qualitative and quantitative data might serve as a starting point for these more complex analyses.
Summary updated