is Differential expression analysis an unsupervised technique?

Question

[Mo]

Hello,

I have been trying to find up/down regulated genes in my data which I was not successful so far. In my previous post, I was told to perform Differential expression analysis using Limma package in R. I found that the function works based on correlation between expression, but it is not very clear how to find the up/down regulated. I searched through web but I could not find a good example which shows how one can perform such analysis!

I am wondering if anyone can explain how to perform DE analysis in R and find the up/down regulated on a set of data where you don't have any extra information about any of genes. I posted an example data set in gist which can be downloaded here.

[iraun]

You maybe want to try DESeq to find differential expressed genes across your samples. I recommend you to follow this manual. It is quite clear, and you can find all the steps explained and also the commands that you need to use in R. Here also there is a nice blog which explains how to use DESeq and take nice plots of expression data.

[Michael]

No, op is working with microarrays!

[Ram]

OP should be adding more relevant tags. I say we abolish the "gene" tag. Most of the stuff in here is obviously related to genes!

[iraun]

Oh! Sorry, I did not read that. Therefore ignore my comment! :)

[Mo]

I could not get the point across, Michael! Can you please explain what is op? Where to find it? Is there any example?

[Ram]

OP = Original Poster = you, Mohammad. You should be adding a lot of relevant details and disambiguation statements in your question. Mentioning that the post is important to you alone will not suffice if the question does not give folks a clear context of your problem setting.

[Mo]

Thanks. For sure, I'll try to be more specific in my other questions. Please let me know if anything is unclear here. I found the admin has changed the tags and added more into it. I did not mean the question is very important only for me, I know many people will face the same problem as I did (this is a very first step in gene analysis to find out which genes are up or which are down regulated) so I think this is an important question. I wrote that I could not perform DE and therefore, I share my problem here to see whether somebody has a solution or has an experience with this type of analysis. I also posted an example data set. I don't know how I can be more specific in this question. Please let me know, I try to explain more.

[Ram]

Thank you.

Being specific helps because you're starting out with data that a lot of people use, methods that a lot of people use, and are looking to get to a result that lot of people also have as their destination.

However you're looking for someone that belongs to all three groups, so if you mention a) microarray, b) Diff Expression analysis and c) list of up-regulated and down-regulated genes, and then explain the problem you're facing, people can tell you how to solve it, and even if you approach is appropriate or needs change.

This has been my experience asking questions on forums.

Answer 1

[Sean Davis]

The Limma user guide is very thorough. However, you'll need to have some basic understanding of statistics, also, in order to best interpret results. To answer your question directly, up-regulated genes (always with respect to a comparison group, of course) have a t-statistic that is and a logFC that is positive while down-regulated genes have the opposite.

As for your implied question, you may want to look at the limma romer, roast, and camera functions, as these perform Gene Set Analysis of various forms.

[Mo]

Thanks for your comment. Actually, I am looking for answers to the point. There are many posts showing what is up or down regulated etc (for example Understanding up and down regulated genes from LOG2 foldchange or foldchange) I did not find any example to show how to perform DE on a microarray data.

To be honest, firstly I was trying to do GSEA but in my previous post, they said GSEA is not the right technique since I did not have a good phynotype question or something like this. I believe all those functions you mentioned are for GSEA which I am not looking for. I simply look to find an example of how to find up/down regulated of a given microarray data.

Answer 2

[andrew]

Hi Mohammad,

I made mention of this in your other thread but I will expand on it here.

Since you are working with Affymetrix CEL files, I may have the shortest path for you without needing to know how to program.

iPathwayGuide has the ability to take Affy CEL files as input. You simply group them into your phenotype groups (e.g. health vs. disease). The application will compute the differential expression and allow you to select cutoffs for significance based on p-value and log2 fold change. In about a month we will also provide additional feedback on QC and normalization. Currently that is being done without feedback to the user. Keep in mind the CEL files can be rather large and it may take several minutes to upload and compute the DE genes.

Then if you want, you can proceed to additional analyses including prediction of miRNAs, pathways, GO terms, and diseases. It's completely free to try with full access to the results for 72 hours.

Here is the list of currently supported Affy CEL files.

Human

• Human Genome U133
• Human Genome U133A 2.0
• Human Genome U133 Plus 2.0
• Human Genome U95
• Human Genome U35K

Mouse

• Mouse Expression Set 430
• Mouse Expression Set 430 2.0
• Mouse Genome 430A 2.0

Rat

• Rat Expression Set 230
• Rat Genome 230 2.0
• Rat Genome U34

[Mo]

Thanks for your message. I am more searching for R or python programs. I am not interested in software or website stuff. However, thanks for your comment. Hope those guys who are interested in software see your message