J express for Protein Microarray Analysis

[Neilrenault]

Hi

I was wondering if anyone has used J Express to analyse Protein Microarray Data.  We are using a protein microarray to profile for immunoglobulin binding using fluroescence.  For each experiment we use a control slide of PBST and a sample slide incubated with sera.  As some proteins autofluroescece it is very important to be able to subtract the control from the sample.  Is there a way I can do this using J Express??  Or is there any other software that I can do this with.

Many Thanks
Neil Renault (neilrenault@hotmail.com)

[Bjarte Dysvik]

We have used J-Express for protein analysis from 2Dgels, but not from protein arrays yet. This should however not be a problem, but if you are just substracting the control (and not using ratios etc.) you probably need a script to do this. In what form are your data now? Do you have one data file for each array or one single tab-delimited file for the experiment? have you imported the data into J-Express?

I can create the script for you.

Bjarte.

[Neilrenault]

Hi Bjarte

Thanks for the reply.  We are using 2 slides.  So we have a data file for each slide.  When we run an experiment we use 2 secondary antibodies with different fluroescent markers (strep cy5 and cy3).  So we use 2 channels, being F635nm and F532nm.  I have managed to produce a table using the two arrays with 4 columns.  Obviously these refer to both the wavelengths (eg. F635nm - B635nm and  F532nm - B532nm) and then for both slides.  We have 3 replicates of all our samples and so it produces this summary table in terms of an average of the replicates.

I agree I think I need a script to enable me to minus the control data from the sample slide.  If you could creat one for me that would be great.  We know J-Express is very powerful and the stats we would like to do are relativily basic for this programme.  Is there a way to tell the programme to look at the SAMPLE slide and express T-test/Standard Deviation / and Standard error values.

The files are spotpix (*.gpr files), do we need to convert them to commo delimited files?

Thanks so much for your help.
Regards
Neil Renault

[Bjarte Dysvik]

Hi Neil,

Just to be sure I understand the layout of your dataset, originally you have 2 slides (both with 3 replicates = 8 arrays) and two channels for each slide. You want to combine the data so that you have for each slide one reported signal being channel1 - channel2?

This should result in a two class dataset with 4 arrays in each class. Then you want statistics such as t-tests between the classes and standard deviation within each class?

is this correct?

J-Express can read the gpr-files directly, so you should not have to create a tabular data file before you put the data into J-Express..

[Neilrenault]

Hi

Just to correct you.  We are profiling for immunoglobulins.  So we run ONE control slide (shows no non-specific binding) and incubate with PBS, then 2 rounds of secondary antibodies and fluroescent markers.  Then with another ONE slide we incubate with sera, the same secondary antibodies and fluroescent markers are then incubated.  So that 532nm for example we detect IgA and at 635nm we detect IgM.  Due to the autofluroescence we need the control to have the same treatment apart from the sera.  All our protein spots are in replicates of 3 on each array, (i.e. 900 protein samples produces 2700 spots)

So for each time we scan one slide we get background and fluroescent values for each spot at both wavelengths.  So what we simply want to do is get the values for the control slide and minus them from the sample slide.  Then do a T-test / Std error analysis for the variation in spot intensity on the sample slide between the 3 replicate spots.

I hope that has made it clearer

Many Thanks
Neil Renault 

[Bjarte Dysvik]

Hi again Neil,

You should first import the data into J-Express by choosing "file" -> "load raw data". Drag and drop the two gpr files into the "experiment design table".
Choose the channel1 and channel2 (F635nm - B635nm and  F532nm - B532nm).
set the "Combine in-array replicates" to NO and make sure "Result data" is set to none.

Then select an array (click an array icon) and click the process tab. Add appropriate filters (for controls etc.) and normalization if needed (e.g. a global lowess normalization). copy the processes to both arrays by clicking the "copy to all" button.

click compile to create a expression matrix.
You should now get a 4 column dataset.

select the dataset and open the script window from the menu "dataset" (select pyton script window)

I created a script which can be found on the forum at: http://molmineus.com/forum/index.php?topic=60.0
copy all script code and paste into the empty script window.
look through the script and see my comments.. verify that the script does what it should.

click the execute button.

If I understood the data layout correctly, the script should produce the results you want.. We can add other statistics later if you want..

Bjarte.

[Neilrenault]

Hi Bjarte

Thanks for the script.  I did all the things you said and I got a very nice table with an average and variance for each of the two wavelengths.  I also got a further 3 columns entitled 'T-statistic'

I manage to copy and paste data into excel and produced a bar graph and sorted them in numerical order.   What I would like to go is add the std error values to this graph.  I have tried plotting this using the 'variance' data but most of the numbers appear not to correlate or are far too big.  The T-statsitic seems to correlate quite nicely.  Should I use to add error bars for std error??

Do you have an suggestions or experience in setting thresholds to saying yes something is significant or no.  I know in DNA arrays expression is quite often occuring or not occuring.  What is the best statistical way to decide on a threshold in the in-between values??

Many Thanks
Neil Renault

[Bjarte Dysvik]

Hi Neil,

I added a second script similar to the first with standard errors added (http://molmineus.com/forum/index.php?topic=60.msg132#msg132).

This script also prints the values of the replicates in the output window so that you can inspect the variation between the replicates manually (or use some excel scripts to calculate other statistics). You can copy these to excel by cltrl-c and ctrl-v into excel.

Setting thresholds for significance depends on a lot of factors such as sample size, tecnical and biological replicates etc. There are a lot of papers written on this issue for microarrays and some theory probably also apply to your protein arrays. I recommend a pubmed search on the issue (some of the big names in this field is probably Kathleen Kerr and Gary chuchill).

best,
Bjarte

[Neilrenault]

Hey Bjarte

I think your scripts are working really well that you wrote for me.  When I paste the data into excel the std error data seems to have little correlation to the data itself.  I think what I need to look at the std error for is to compare the variation between the fluroescence of the sample spots. I'm not quite sure why it is doing this but my higher positive samples have a final fluroescence value of 150-600.  However the std errors keep working out to be anything from zero to 57 million.  Any ideas what could be going wrong here?  I could send you the gpr files if you like of the two arrays that I am carrying out this test analysis on.

Many Thanks

Neil

[Bjarte Dysvik]

Hi Neil,

Are you using the std. error values to call proteins as differentially expressed?
I think there are nothing going wrong, but the large difference in expression (especially for more intense spots) are affecting the variance and standard errors.. Perhaps you sould consider calculating a fold change between the two samples instead? Or ANOVA (or even SAM?)

Bjarte.

[Neilrenault]

Hi Bjarte

Our protein microarray simply has immobolised protein spots on them.  To this we then measure antibody binding by detecting fluroescence via secondary antibodies.  In other words the higher the fluroescence of the spot the more immunoglobulins have been detected.  We are not measuring differences in expression.  When you look at most of the spots they appear very similar.  Indeed when you look at the raw data values of the sample spots the values for each replicate may be 140/130/160.  So we would not expect the std error to be massive.  Even if one of these 3 values was 1400 it would not be 57 million.

Previously when we have worked out std error we have used the following formula.

Std Error = Standard Deviation of 3 replicate spots
                  SQRT √ (of the Number of data values i.e. 3)

Would it be possible to alter the script so it puts these results in the results table.

Thanks again
Neil

[Bjarte Dysvik]

Hi Neil,

Of course you are right.. There was a power instead of a square-root in the standard error calculation.. I have updated the script, please try again.

best,
Bjarte.

[Neilrenault]

Hi Bjarte

The good news is that the script seems to work pefectly, thanks very much.  From the data table that is produced I have copied this to excel to produce a graph with my samples in ascending order of intensity with Std error values.  I assume there is no way to plot a graph like this in J Express.  Obviously the ideal would be to be able to plot a graph with some error values and link this to example spots of the control/sample slides.  I can't think how else I could present this data.  I know J Express has some graphical statistical tools.  Having a way to visually compare the spots, sorted according to intensity,  would be really useful for our analysis.

Thanks alot

Neil

[Bjarte Dysvik]

Hi Neil,

I think the best way to do this is to copy the identifiers for the proteis you are interrested in and paste them into the search field of the search and sort window in J-Exress (create a selection by clicking the red asterisk). This way you can create a selection of proteins to do further work on. If you added the jpg-images in the SpotPix suite window (where you loaded the gpr files), you can click the view image spots from the dataset menu and view the actual spots that corresponds to the selected proteins. Have a look at the two tutorial documents in the learning section for examples on how to do this.

There are many charts available if you can use the scripting functionality, and I would very much like to help you on this. But unfortunatly I have too many other things to do at the moment.

Working with Excel together with J-Express is a good approach to your analysis.. Most of the functionality present in J-Express are implemented for streamlined microarray analysis with biological and technical (array) replicates, but much of the preprocessing and selection functionality should fit the data you have.

good luck with further analysis 

best,
Bjarte.

[Neilrenault]

Hi Bjarte

Thanks for the scripting, it seems to work really well.  I have been looking into the 't-statistic' that is produced by the script you wrote for me.  Obviously this allows the program to conduct a simple T-test, which you can compare to critical values to show significance.  Could I possibly ask you to what does this t-statistic refer to??  Does it show that there is a signifcant difference between the data of the control slide for a specific spot type and the patient slide.  Or does it just show a difference between the 3 replicate spots within each of the 2 slides.

I am sorry to trouble you again.
Many Thanks
Neil