Differential Analysis Of Proteomics Data

Question

[Assa Yeroslaviz]

Hi,

I have three different proteomics data sets, each of them in duplicates. Regardless of the problem not having triplicates for statistical power. I wpuld like to know in general how I can analyse the differential expression (quantitative analysis) of the two conditions in each of the runs.

One of the experiments is SILAC, the other two are label-free. I would like for each run to compare the two condition to check for statistically significant changes.

My idea was something bases on the (pair-wise) analysis of microarrays, but I am not sure exactly (if at all needed) how to normalize the data sets.

The data I have is the output from the maxQuant (excel sheets with loads of columns, in some of them it says Ratio H/L normalized). Does it means I don't need to normalize the data again?

thanks for any help Assa

Answer 1

[Ahill]

For the SILAC data, the "Ratio H/L" column will contain the ratios of the "heavy" (labeled) sample to the "light" (unlabeled) sample. The SILAC ratios do need to be normalized. Judging from the name of the column you mention, they may already be so. To confirm that you should check that the log-ratios in your dataset are centered on zero. To do differential expression on the SILAC data, assuming your ratios are at the peptide (not protein) level, one approach is to aggregate the log-ratios for all peptides belonging to each protein, and then use, for example, a one-sample T-test to compare those log ratios against a null hypothesis of zero (no differential expression). Be aware of some technical issues that can affect SILAC quantitation, like proline conversion [1].

[Assa Yeroslaviz]

yes, as far I understood it from the Biologist who gave me the data, they are LFQ-normalized (label free quantification), using the maxquant software. I am not exactly sure what it means, as I am not familiar with the softeware and the Bioloist couldn't really explain it. Does a one-sample T-test is really meaningful when i have no replica? I mean I will calculate the ratio between the two conditions of the H/L-ratios, but can I assign any statistical significance to the experiment without replica? thanks for the info.

Answer 2

[Charles Warden]

Maybe try DanteR? Haven't tried it myself - let me know how things turn out.

At least from a theoretical standpoint, you can try reading about some of the algorithms I list under the Mass Spectrometry section:

http://cdwscience.blogspot.com/2013/03/bioinformatics-101-protein-analysis.html

[Assa Yeroslaviz]

I would like to try it, but can't figure how to use it. It was difficult enough to install it, after I have found out that I need the newer Gtk2 version from here. I than installed everything according to the instructions, but it just doesn't stat. The R session just quit with a strange error massage:

Warning messages:
1: package ‘gWidgetsRGtk2’ was built under R version 3.0.2 
2: package ‘RODBC’ was built under R version 3.0.2 
3: package ‘plotrix’ was built under R version 3.0.2 
4: package ‘nlme’ was built under R version 3.0.2 
5: package ‘scatterplot3d’ was built under R version 3.0.2 
Options file does not exist
DanteR, v 0.2
> R(19961,0x7fff7b69f960) malloc: *** error for object 0x4024000000000000: pointer being freed was not allocated
*** set a breakpoint in malloc_error_break to debug
Abort trap: 6

It seems that the package looks for some kind of options file, but there none in the installation. Do you have any idea what does suppose to be?

[Charles Warden]

I put together that list of software when I was expecting to assist with some mass spec protein analysis, but I never ended up getting a chance to test those tools. So, unfortunately I can't be much more help.

Maybe you can try contacting the corresponding author on one of the papers:

http://bioinformatics.oxfordjournals.org/content/24/13/1556.long

http://bioinformatics.oxfordjournals.org/content/28/18/2404.short