Using Pre-existing Microarray Datasets to Increase Experimental Power: Application to Insulin Resistance

Bernie J. Daigle, Alicia Deng, Tracey McLaughlin, Samuel W. Cushman, Margaret C. Cam, Gerald M. Reaven, Philip S. Tsao, Russ B. AltmanPLoS 2010 · DOI

Abstract

DNA microarrays are powerful tools for identifying differentially expressed (DE) genes but are notoriously susceptible to experimental noise. To mitigate this, researchers typically rely on replicates, which can be prohibitively expensive or impossible in resource-limited settings. We present the SVD Augmented Gene expression Analysis Tool (SAGAT), a mathematically principled approach that leverages publicly available microarray data to reduce uncertainty in gene expression measurements. By projecting new data onto "eigengenes" derived from thousands of public samples, SAGAT increases experimental power, effectively adding up to 2.72 arrays to an experiment's sample size for free. Applied to a study on insulin resistance, SAGAT increased the detection of significant genes by 50%, revealing key pathways in inflammation and fatty acid synthesis.

Introduction: The Challenge of Experimental Noise

In genomics, the high-throughput nature of DNA microarrays is often undermined by noise. While increasing biological replicates is the standard solution, it is often impractical in clinical studies. Meanwhile, hundreds of thousands of microarray experiments reside in public repositories like the Gene Expression Omnibus (GEO). These datasets are typically ignored because their specific conditions—different tissues, diseases, or laboratories—do not match the new experiment. SAGAT addresses this by extracting robust biological signals from this "messy" prior knowledge to improve the accuracy of specific, small-scale studies.

Methodology: The SAGAT Framework

SAGAT uses Singular Value Decomposition (SVD) and pseudoinverse projection to bridge the gap between public knowledge and new experimental data.

  1. Extracting Modules: A massive compendium of public data (over 30,000 arrays) is decomposed into "eigengenes." These are mathematical representations of transcriptional modules—groups of genes that consistently move together across many different conditions.
  2. Projection and Smoothing: New experimental data is projected into this "eigengene space." Because these eigengenes represent robust biological patterns seen across thousands of samples, the projection helps "smooth out" noise unique to the small experiment.
  3. The SAGAT Score: The tool calculates a score (similar to a t-statistic) that incorporates the variance information from these shared modules, making it easier to identify DE genes even with very few replicates.
Figure 4: Gene-gene coexpression matrix derived from public microarray data compendia.

Key Insights: Biological Reality in Mathematics

The authors found that the top eigengenes produced by SVD map directly to fundamental biological processes such as translation, cell cycle regulation, and immune response. Crucially, genes that are differentially expressed in a disease state are significantly more likely to belong to these coexpression modules than genes that are not. This "joining" behavior allows SAGAT to boost the signal of genes that behave like known biological pathways while suppressing random experimental artifacts.

Results and Validation

The effectiveness of SAGAT was demonstrated across several benchmarks:

  • Sample Size Boost: In validation tests using "gold standard" cancer datasets, using SAGAT was equivalent to adding between 1 and 2.72 physical arrays to the experiment.
  • Insulin Resistance Study: When applied to a dataset of 24 women matched for obesity, SAGAT identified 18 significant genes compared to 12 identified by traditional methods—a 50% increase.
  • Experimental Confirmation: qPCR was used to verify 11 of the genes identified by SAGAT. All 11 matched the predicted direction of change, and several reached statistical significance that was missed by traditional microarray analysis.
  • Biological Discovery: SAGAT revealed a coherent biological narrative for insulin resistance involving inflammation (CXCR4), cell differentiation (FOSB), and fatty acid synthesis (FASN, ELOVL6), suggesting that obesity-independent insulin resistance is driven by specific failures in lipid management and inflammatory response.

Conclusion and Future Directions

SAGAT demonstrates that public data contains structural information that can significantly enhance new experiments, even when the data sources are unrelated. While currently platform-specific and based on linear assumptions, the method provides a powerful, free tool for maximizing biological discovery from subtle transcriptional responses. Future work includes adapting the framework for RNA-seq data and exploring non-linear manifold learning techniques to capture even more complex biological signals.

Summary updated

Abstract

Although they have become a widely used experimental technique for identifying differentially expressed (DE) genes, DNA microarrays are notorious for generating noisy data. A common strategy for mitigating the effects of noise is to perform many experimental replicates. This approach is often costly and sometimes impossible given limited resources; thus, analytical methods are needed which increase accuracy at no additional cost. One inexpensive source of microarray replicates comes from prior work: to date, data from hundreds of thousands of microarray experiments are in the public domain. Although these data assay a wide range of conditions, they cannot be used directly to inform any particular experiment and are thus ignored by most DE gene methods. We present the SVD Augmented Gene expression Analysis Tool (SAGAT), a mathematically principled, data-driven approach for identifying DE genes. SAGAT increases the power of a microarray experiment by using observed coexpression relationships from publicly available microarray datasets to reduce uncertainty in individual genes' expression measurements. We tested the method on three well-replicated human microarray datasets and demonstrate that use of SAGAT increased effective sample sizes by as many as 2.72 arrays. We applied SAGAT to unpublished data from a microarray study investigating transcriptional responses to insulin resistance, resulting in a 50% increase in the number of significant genes detected. We evaluated 11 (58%) of these genes experimentally using qPCR, confirming the directions of expression change for all 11 and statistical significance for three. Use of SAGAT revealed coherent biological changes in three pathways: inflammation, differentiation, and fatty acid synthesis, furthering our molecular understanding of a type 2 diabetes risk factor. We envision SAGAT as a means to maximize the potential for biological discovery from subtle transcriptional responses, and we provide it as a freely available software package that is immediately applicable to any human microarray study.
BibTeX
@inproceedings{daigle2010using,
  author = {Bernie J. Daigle and Alicia Deng and Tracey McLaughlin and Samuel W. Cushman and Margaret C. Cam and Gerald M. Reaven and Philip S. Tsao and Russ B. Altman},
  title = {{Using Pre-existing Microarray Datasets to Increase Experimental Power: Application to Insulin Resistance}},
  booktitle = {PLoS},
  year = {2010},
  doi = {10.1371/journal.pcbi.1000718},
}
Figure