Just like the bagel, microarray experiments have their own subtleties and nuances that can make even "identical" experiments different. But when it comes to biomedical research, there's more at risk than disappointed taste buds. A researcher might use arrays to identify a gene expression profile for early cancer detection, but if a clinical laboratory can't reliably repeat that experiment, the ground-breaking study will never help a soul.

Creating data that can be shared, compared, and developed by the worldwide scientific community requires standard controls and "best practices" for microarray data generation and interpretation. Towards this goal, a group of academic and industrial scientists have formed The Tumor Analysis Best Practices Working Group. The organization's objective is to identify areas of variation between array experiments, and to help develop standardized steps for ensuring consistent and reproducible research, including tumor analysis studies in clinical trials.

The Working Group published their recommendations in the March issue of Nature Reviews Genetics. While every array platform requires technology-specific guidelines, the group focused on developing best practices for Affymetrix GeneChip® brand microarrays because of their widespread use in laboratory research and clinical trials and the already standardized manufacturing process. They concluded that the primary areas for microarray standardization include experimental design, technical variability, and data analysis and interpretation. The Working Group also discussed the efforts underway to establish data-reporting guidelines.

Experimental Design
The critical aspects of successful experimental design defined by the Working Group were three-fold: 1) using sufficient biological replicates, 2) making comparisons between equivalent tissue types, and 3) standardizing tissue sampling and storage procedures. The payoff of adhering to these accepted best practices is high-quality data that can readily be compared between different studies, sites, and dates—a more efficient and effective use of resources and precious patient samples.

Technical Variability
The quality and quantity of RNA examined in a microarray experiment is often a critical source of variation as well. As with our bagel-baking metaphor, using varying amounts of different quality ingredients inevitably yields varying results. With microarrays, the main ingredient is RNA. To generate the same quality product again and again, the Working Group calls for standard RNA concentrations and quality control practices to be used in all microarray experiments.

"Adherence to quality control criteria, using standard RNA isolation and processing methods, should yield data that are consistent between laboratories and intrinsically comparable," says Janet A. Warrington, Ph.D., Vice President of Molecular Diagnostics Research & Development at Affymetrix, and a member of the Working Group. "The same set of criteria can also be used as best practices for data generation in the design and conduct of clinical trials."

Data Analysis and Interpretation
Twenty-two different probes are used to measure the expression level of each transcript on Affymetrix® expression arrays. And depending on the way those 22 probes are evaluated to create a single measure of gene expression, the same exact microarray can result in very different final data. The numerous statistical programs now available to analyze and interpret microarray data are perhaps the greatest sources of confusion between array experiments.

"Members of the Working Group encountered up to 50 percent variation between comparisons of two different algorithms," explains Warrington. "The Working Group recommends using multiple methods for comparison and prioritization of expression patterns for identification of robust classifiers." While it is difficult to standardize data analysis methods that are continually improving, the need for consistent analysis is as important as standard experimental design and technical processes.

Microarrays have been used to identify specific gene expression profiles to predict, diagnose, or classify disease from a patient sample. Developing standardized ways to compare gene expression from one array to another is essential to this process. However, there are numerous statistical and bioinformatics methods available to compare array data generated by probe-level analysis, and different programs often provide different results. The Working Group recommends that users experiment with various statistical methods to develop the most accurate predictions.

"It is in this realm that microarray bioinformatics becomes avante-garde, and with the ground-breaking nature of research comes considerable debate as to what is appropriate in any specific situation," says Warrington. "Inclusion of a seasoned statistician in developing experimental design, sample numbers to adequately power your study, and, when possible, prospective validation of any findings using new patient samples is the acid test of predictive performance."

Data Reporting Guidelines
The final step in array experimentation—reporting final data—also requires standardization, enabling scientists to properly compare data from different experiments. The Microarray Gene Expression Data Society recently developed such data-reporting guidelines, but the guidelines do not address essential issues of variability in data generation and interpretation discussed by The Tumor Analysis Best Practices Working Group.

By developing best practices for the complete microarray workflow, array analysis can be performed according to defined standards and regulations that will be necessary for clinical applications. Microarrays are currently being used in over 40 clinical trials, but wide use of the technology in the routine setting of clinical care will require additional standardization.

And whether you're performing array experiments or making bagels, establishing standards and "best practices" remains the best way to ensure quality and consistent results. Ironically, we seem further along with the microarray than we do with the bagel.

Find out more about microarray standards contols and best practices.