• Basic training on any of the software packages.
• Data Processing: QC reports with remarks. Standard outputs: Robust increasers/decreasers, Scatter plot w/ fold change lines, Lists of genes up/down 2-fold, 2.5-fold, and 3-fold.
• Optional: Time series data processing.

I. Experimental Design

It is imperative that microarray experiments be designed carefully to minimize sources of variation and to produce meaningful data[11].  While it is expensive to run multiple chips for biological replicates, this approach is necessary to be able to estimate and account for biological variability and interpret results with any degree of confidence.  For experiments that are exploratory in nature, RT-PCR or Northern blotting is often used to confirm the results.  In addition, it is important to validate the results using statistical approaches [1].

For initial microarray experiments, a pilot study for testing a single hypothesis using 3 or 4 biological replicates is recommended [1,2].   From the pilot study the estimated variance of gene expression can be calculated for use in determining the number of replicates needed in subsequent related experiments.  In such subsequent experiments, the hypothesis may be refined toward a more specific goal.

BCF Services: 

• BCF staff will meet with you in conjunction with members of the CCIT Statistics team to discuss experimental design, including the number of replicates needed in order to derive statistically significant conclusions from your data.  No cost.

II. Processing Array Data

There are many software packages available, both commercial and free. Most are not terribly easy to use. A feature summary table is provided below.  Commercial packages have been licensed by BCF and may be used only on the machines in the computing lab.  BCF staff can provide training on many of the popular software packages.  It is possible to process your data with more than one software package and compare the results.  To apply for an account on the BCF MicroArray Analysis PCs, go to our Accounts page.

The steps typically used in processing raw Affymetrix data include the following:

Calculate signal -> Normalize -> Compute Fold Change; Compare Chips > Clustering or Statistical Analysis

Software that can process raw Affy microarray data must calculate a signal intensity value for each probe set.  Affymetrix MAS 5.0 software uses a Tukey's Biweight Algorithm to calculate the average signal for a probe set [3].   The free software dChip [4] uses a model-based analysis, which yields less noise in the lower expression levels (a significant improvement over the Affymetrix signal calculations).  Other free packages (BioConductor/ RMAExpress) use a robust multi-array average, which reduces noise even more than dChip [5, 6, 7].

Several software packages provide clustering and/or Statistical Analysis tools. For a good introduction to clustering techniques see [8].  The BRB ArrayTools [9] package can be used as a plug-in to Excel, an advantage for those familiar with Excel.  If you are interested in the MIAME standards required by many journals, read [10].

MicroArray Software Features:

BCF Services (see Costs below):

• Basic training on any of the above software packages.
• Data Processing: QC reports with remarks. Standard outputs: Robust increasers/decreasers, Scatter plot w/ fold change lines, Lists of genes up/down 2-fold, 2.5-fold, and 3-fold.
• Optional: Genespring outputs. Time series data processing.

III. Costs for BCF Services

One hour of training on the software or data processing is provided free of charge.  Additional staff hours are billable (at approximately $40 per hour.)

IV. References   (**) Recommended Reading

1. Statistical issues with microarrays: processing and analysis.  Robert Nadon and Jennifer Shoemaker, TRENDS in Genetics, Vol.18 No.5, May 2002.(**)
2. GeneChip® Expression Analysis: Experimental Design, Statistical Analysis, and Biological Interpretation, Notes from Affymetrix Data Analysis II course, Spring 2003.
3. Affymetrix GeneChip Expression Analysis Data Fundamentals Guide. http://www.affymetrix.com/support/technical/index.affx(**)
4. Model-based analysis of oligonucleotide arrays: model validation, design issues and standard error application, Cheng Li and Wing Hung Wong, Genome Biology  2(8): research0032.1-0032.11, 2001.
5. Summaries of Affymetrix GeneChip probe level data, Izarry et. al., Nucleic Acids Research, Vol.31, No.4, 2003.(**)
6. A Comparison of Normalization Methods for High Density Oligonucleotide Array Data Based on Bias and Variance, Bolstad, B.M., Irizarry R. A., Astrand, M., and Speed, T.P., Bioinformatics 19(2):185-193, 2003.
7. Computational Analysis of Microarray Data, John Quackenbush, Nature Reviews Genetics vol. 2, June 2001.(**)
8. BRB-ArrayTools Version 3.0 User's Manual, Dr. Richard Simon, Biometrics Research Branch, NCI and Amy Lam, The EMMES Corporation. http://linus.nci.nih.gov/pilot/index.html
9. Conforming to MIAME a guide to prospective authors of Affymetrix-based microarray papers, Miller, et. al., Paterson Institute For Cancer Research, Manchester, UK. 
10. Experimental Design of DNA Microarray Experiments, R. Simon, K. Dobbin, Biotechniques, Mar 2003.