Determing top differential expression of proteins using non-microarray dataset

Question

[mllerch]

I have a protein expression data set using a novel platform and would like to determine the differential expression of the proteins between two tissue types. There are 8 independent samples for each condition. I have 858 proteins and would like to narrow it down to a manageable and meaningful heatmap. I am not a biostatistician but I am learning R. It seems like the limma package in R can do what I want, but my data is not in the same format. I have already normalized my data. Does anyone have some advice for how to analyze my data to give a similar output?

fit <- eBayes(lmFit(eset,design))

> topTable(fit, coef=2)
              ID         M        A         t      P.Value        B
1016     1914_at -3.076231 4.611284 -27.49860 5.887581e-27 56.32653
7884    37809_at -3.971906 4.864721 -19.75478 1.304570e-20 44.23832
6939    36873_at -3.391662 4.284529 -19.61497 1.768670e-20 43.97298
10865   40763_at -3.086992 3.474092 -17.00739 7.188381e-18 38.64615
4250    34210_at  3.618194 8.438482  15.45655 3.545401e-16 35.10692
11556   41448_at -2.500488 3.733012 -14.83924 1.802456e-15 33.61391
3389    33358_at -2.269730 5.191015 -12.96398 3.329289e-13 28.76471
8054    37978_at -1.036051 6.937965 -10.48777 6.468996e-10 21.60216
10579 40480_s_at  1.844998 7.826900  10.38214 9.092033e-10 21.27732
330      1307_at  1.583904 4.638885  10.25731 1.361875e-09 20.89145

[mllerch]

While I did not find the user guide entirely useful. This post was a good guide. Limma calls all genes as differentially expressed - what am I doing wrong?

Answer 1

[Sean Davis]

Limma is not specific to microarrays and can take a matrix as input, which you can likely get from your normalized expression values. The design matrix will involve a single factor, so you can follow the workflow in the Limma User Guide for single-factor designs.

[mllerch]

I have looked at the User Guide and have not found it very useful for formatting my data. I do not know what my matrix should contain. I am simply comparing 2 conditions like WT and Mut, but I have 8 patient samples with both Tumor and Normal tissue.

Specifically what should the eset file look like? Do I need specific names for each column? Right now, I have all the data in a single file (as below) or divided by pt.

head(df13)
GeneID pt.num       D       T id  RatioT.D      log2T.D
1    A2M      1  8876.5  8857.9  1 0.9979046 -0.003026224
2   ABL1      1  2120.8  1664.9  2 0.7850339 -0.349173049
3   ACP1      1  1266.6  1347.1  3 1.0635560  0.088895967
4   ACP5      1 67797.6 24218.2  4 0.3572132 -1.485142562
5 ACVRL1      1   650.1   822.8  5 1.2656514  0.339880140
6   ACY1      1  6264.8  7112.9  6 1.1353754  0.183169428

head(df18)
     row.names    log2NormT.D.2    log2NormT.D.4    log2NormT.D.5    log2NormT.D.6    log2NormT.D.7    log2NormT.D.8
1    A2M    -0.221639704    -0.572627315    -1.7253267763    -0.4876931057    -0.7466838821    -1.300579696
2    ACP5    -3.375111627    -3.558914753    -5.7284910478    -2.4661145121    -2.4836598593    -2.729808764
3    ADIPOQ    -1.447350719    -0.407452553    -0.8710711001    -1.2626372438    -1.0012156238    -1.080200126

Answer 2

[pld]

I've used this before with Kinome and Full Moon arrays: http://saphire.usask.ca/saphire/piika/

So long as your inputs are correct, Piika doesn't need to know any layouts outer than duplicate probes. I think the only major difference between something like this and Limma is Piika uses VSN to normalize and automates creating several plots.

You can download the script if you would rather run it locally, it can be a bit easier if you have lots of things to analyze.