Microarray has various uses, and to analyse the data a main function classification is used. There are many methods used to classify the data but what are the best and most frequently used methods? Are there any well-known, web-based tools for microarray analysis? Where can I find datasets of gene expression profiles, where I can find and understand the matrix of the dataset?

  • $\begingroup$ try the bioconductor package. $\endgroup$ – WYSIWYG Oct 26 '13 at 6:15
  • $\begingroup$ Sebida and DGRP have data available $\endgroup$ – rg255 Oct 26 '13 at 7:17
  • $\begingroup$ Come to think of it most or all gene expression data has to be deposited online - try looking at the gene expression omnibus. And I echo recommending bio conductor (in R) $\endgroup$ – rg255 Oct 26 '13 at 7:20

This is a fairly broad question and a book or two will give a more complete answer. I'm assuming you are interested in expression microarrays. Genotyping microarrays are also very popular these days and are quite different though.

Where can I understand the matrix of the dataset ? This is a little unclear as a question, but I'll start with a numerical description. Microarray data is reducible to a one dimensional array of intensities. As a vector of numbers, each representing the amount of a given species of RNA in a sample, it has many interpretations. Most of what is said here applies to RNASeq data too.

These numbers are sequence dependent - that is, they should not be interpreted as an absolute count of a specific mRNA in the sample, so comparing two entries in the vector (e.g. "gene A has 10% more RNA than gene B") will probably not be useful.

For this reason, microarray experiments are typically done as difference experiments, where two or more conditions are compared, producing a matrix of numbers, where N rows represent N experiments, and the M columns each represent a probe set / gene. The transpose of this matrix is also common.

**Methods to classify microarray results. ** Taking the matrix of data, there are many ways of interpreting data in matrix form and most of them have been tried credibly with microarray data.

The most common are to take a set of probe sets whose biological action are related (e.g. all in a metabolic pathway and display them as a heat map where the columns have been clustered by a distance based algorithm to produce a tree, where neighbors have the greatest numerical similarity.

heat map

For purely numerical methods, Principal Component Analysis is often used (despite the fact that this is not a great method), but Self Organizing Maps, neural nets and just about any machine learning algorithm have been used.

Most common method by far though, is Analysis of Variants (ANOVA), where the probe sets/ genes have been scored by their greatest to smallest variance. The most common case of this is a two condition difference experiment where the largest differences are taken as results. This is not a great experiment, but the data is not super cheap to produce.

**Microarray data repositories ** Gene Expression Omnibus (GEO) is a highly regarded repository for microarray data, with over a million sets of data. They also have some tools, see below.

EMBL ArrayExpress claims to have slightly more data than GEO (1.2 M assays).

There are many many specialized microarray websites, mostly focusing on a particular organism (E coli for instance) or problem (like cancer). Google is pretty helpful for this sort of thing once you have the right keywords.

**Web based tools for microarray analysis ** There are several tools you will use for microarray analysis and there are lots of websites for different aspects of this data. I'm going to cite a couple, but the most commonly used tools are command line tools like R/bioconductor or the software that comes from the array manufacturer.

GEO2R is a great tool to browse expression data experiments. It does a simple version of ANOVA and will really help you find some cool biology quickly.

VAMPIRE is a web analysis suite for microarray analysis. Scaling the data into numbers from images to intensities to data analysis to compare several arrays. Its probably one of the most comprehensive analysis sites out there.

  • $\begingroup$ "Principal Component Analysis" not "Principle Component Analysis". It analyses principal components. Misspelling it may demonstrate why you find it a problem, because you need to understand the principles behind the principals. $\endgroup$ – Blessed Geek Oct 27 '13 at 8:11
  • $\begingroup$ I don't see why you would not be predisposed to using PCA but yet pre-disposed to accepting proximity clustering and self-org maps. It doesn't matter which chicken lays the egg. They are chicken eggs. $\endgroup$ – Blessed Geek Oct 27 '13 at 21:22
  • $\begingroup$ I don't it is a discussion we should indulge in, in this comments section - but in short, the people using it (in the link you provided) may not understand the meaning of "principal components". $\endgroup$ – Blessed Geek Oct 28 '13 at 0:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.