I got a request to evaluate some software by Axon
Instruments for the analysis of microarray data. The software,
Acuity version 3.1, costs $4,000 per person
and has to compete with other commercial software such as
StatSci.org has a nice
list of companies and institutions
that produce statistical analysis software for microarrays as part of their
overview of microarray data analysis.
Perhaps the most interesting product of the bunch is
Bioconductor, an open source library for the
R programming language. Both Bioconductor and R are
distributed under the GNU General Public
License, which allows you to modify the source code as long as you agree that if you
distribute your modifications, you distribute them with the source code and allow others to
make further modifications. This requirement is called "copyleft" to distinguish it from
"copyright."
I wrote a
brief introduction to the R language in PDF format. There are lots of
other tutorials for R and Bioconductor
that are better written than mine. In particular, the "Overview
of R and Bioconductor" handout in PDF format is a nice place to start.
R (and Bioconductor) is an object
oriented language, which makes it both difficult to learn and very powerful. Microarray
data is stored in a class called exprSet. This class includes information about expression
levels, standard errors of expression levels, phenotypic data, annotation, description, and
notes.
Phenotypic data represents information about the individual or cell line that produced
this particular microarray, such as the gender of the patient, the tissue being sampled, and
so forth. You can get the pheontypic data from a exprSet class by using the pData() function.
The hardest thing about microarrays is that there are just so many genes that you can
evaluate the expression levels for. Bioconductor allows you to annotate these genes. For
example, if you ran an experiment with the Affymettrix Hu6800 chip, your data set will have
the Affymettrix proprietary names for the genes. Bioconductor has a library that allows you
to find out the symbol, genename, locusid, and other information using the Affymettrix name.
You can also get the PMIDs (PubMed Identification) for papers that cite the particular gene
of interest.
I will write up more details about Bioconductor when I get the time and energy.
In contrast to Bioconductor, Acuity is menu driven, so you just point and click to get a
data analysis. It sounds easier, but quite honestly, you have to work just as hard with
either program because there is no such thing as an easy analysis for microarray data. The
acuity menu has 9 menus:
1. Import Data
-
Import Microarrays
-
Import Substance Annotations
-
Import Microarray Parameters
2. Organize Microarrays into Folders
-
New Folder
-
Select microarrays in the tree
-
Drag microarrays into new folder
3. View Data
-
Select microarrays in the tree
-
Open Selected
-
View individual feature values
-
Change the data type displayed
-
View microarray parameters
4. View Statistics
-
Select Microarrays
-
View Statistics
5. Normalize Microarrays
-
Select microarrays in the tree
-
Open Normalization Wizard
6. Create a Dataset for Analysis
-
Select microarrays in the tree
-
Create and Open Data Set
7. Manipulate a Dataset
-
Sort Columns
-
Average Columns
-
Normalize to Column
-
Remove Selected Row
-
Remove Selected Columns
8. Cluster a Dataset
-
Hierarchical Clustering
-
K-Means Clustering
-
K-Medians Clustering
-
Gap Statistic
-
Self-Organizing Maps
-
Gene Shaving
-
Principal Components Analysis
9. Advanced Tips
-
Using the Query Wizard
-
Filtering a dataset
-
Acuity with multiple users
Notice that most of the menus deal with data management. This is not surprising, since the
size of a typical microarray experiment makes data management quite difficult.
As far as I could tell, Acuity could not analyze data from gene chips (Affymetrix
chips). Version 4.0 of the software, I'm told, will have support for gene chip data.
The normalization features of Acuity look quite good and is very easy to use. The problem,
of course, is that there are a lot of choices for normalization, and the typical user won't
know what choices are best. That's not something that you can really blame Acuity for, any
more than you can blame Ford Motor Company because some of its cars are driven by 85 year
olds who are half blind. When I find the time, I will write a short web page about
normalization.
The disappointing aspect of this software is that all of its methods fit into a category
of analysis called unsupervised learning or class discovery. I did not find any methods in
the category of supervised learning (class prediction).
If you want to identify genes that have differential expression across two groups of
microarrays, Acuity offers you either a t-test or the Mann-Whitney test. The typical
microarray has thousands of genes, so some sort of adjustment is needed for the p-values for
these tests. Acuity offers the Bonferroni and Benjamini-Hochberg corrections for multiple
comparisons.
There are several levels of software integration which add to the functionality of this
program.
-
Acuity comes with an industrial-strength database,
Microsoft SQL Server, that allows several people to use the same system and to control
the extent to which information is shared.
-
Acuity also bundles in a nice little program called
pdfFactory that
allows you to convert output from this software into PDF format.
-
Acuity is integrated with another Axon Laboratories product,
GenePix Pro, which does image
analysis of individual spots on a microarray experiment.
-
The Acuity system also produces niche html output and has web scripting features.
The Help files are quite good. There were a few statements that were poorly or ambiguously
worded, but for something as complex as microarray analysis, some imperfections are to be
expected. I did notice some stray entries in the keyword index (<invalid> and <no data>).
Also the entry under "Calculate Multiple Groups" was blank.
