BookmarkSubscribeRSS Feed
TRR21

Community Trekker

Joined:

Nov 4, 2016

Eigengene values from a set of gene expression values

Hi 

In a gene expression data set (1000s of genes expression values for 100s of samples), I would like to get an eigengene value using values for a specific set of genes for each sample (ie like a gene signature score) — expression values for different genes are in wildly different scales, with some values missing (not detectable - may be imputed with 0).

 

Is there a way to accomplish this in JMP?

 

Thanks!

6 REPLIES
chris_kirchberg

Joined:

May 28, 2014

Re: Eigengene values from a set of gene expression values

Hi @TRR21

Can you tell us more about what you are trying to do with the eigengene values? Is it for normalization, for grouping similar expression patterns?

PCA, MDS and Clustering all are methods that one could derive an eigengene value and available in the Multivariate or Clustering sections section of JMP 14.

There are a few ways one could create eigengenes, but without more information about your workflow or the methods you are looking for, it is difficult to guide you to the appropriate place.

Best,

Chris


@TRR21 wrote:

Hi 

In a gene expression data set (1000s of genes expression values for 100s of samples), I would like to get an eigengene value using values for a specific set of genes for each sample (ie like a gene signature score) — expression values for different genes are in wildly different scales, with some values missing (not detectable - may be imputed with 0).

 

Is there a way to accomplish this in JMP?

 

Thanks!


 

TRR21

Community Trekker

Joined:

Nov 4, 2016

Re: Eigengene values from a set of gene expression values

I am simply trying to aggregate the signals from a set of gene expression values (based on a priori hypothesis) into a single continuously distributed score, which I can use to correlate with other attributes (eg. how does score high vs score low samples compare for X, Y and Z), formulate new hypotheses and test that signature in independent experiment(s). What I have is JMP12 as of now.
chris_kirchberg

Joined:

May 28, 2014

Re: Eigengene values from a set of gene expression values

Thanks for the clarification.  You can use PCA to get the PCs that can then be used in Fit Model for analysis with X, Y or Z as responses (Or visa versa).  You could also cluster the genes to get the cluster numbers and then perform some aggregation/standardization of the genes in each cluster then use that value for each cluster to use Fit Model, much the same way as using PCs. Hieracrhical Clustering in JMP 12 allows one to save out the cluster means and options to standardize the data before clustering.

 

Another thought is potentially using PLS (partial least squares) where your X, Y and or Zs can be responses and genes as factors. This way you dont have to aggregate the genes, but only look for ones that have an impact on X, Y and/or Z and then fit a model for those that have an impact.

All of these are found in the multivariate section within Analyze menu for JMP 12.

There really are not any explicit eigengene scored methods in JMP other than using the above standard tools to get to the end.

 

Does that help explain what is available out of the box in JMP 12?

 

Chris

 

TRR21

Community Trekker

Joined:

Nov 4, 2016

Re: Eigengene values from a set of gene expression values

Thanks, Chris. As a relative newbie, I am making sure I get this right. I guess hierarchial clustering and PLS are for discovering a novel gene sets? Those are helpful pointers for discovery, but my question was about a predefined set of genes with the whole dataset.

From what I understand, I can just run PCA for my genes of interest, and get the prin1 and this should be analogous to the eigengene value I am looking for?
chris_kirchberg

Joined:

May 28, 2014

Re: Eigengene values from a set of gene expression values

Yes, Hierachical Clustering and PLS (and even PCA for that matter) can help reduce dimentionality of the data (limit the sets of genes) to find or discover set (or sets).

 

So you already have the sets of genes defined as sets within the data table.  Do you have your genes as rows or columns? How have you desginated the sets of genes in the table?

 

PCA on the genes (as columns) could be used. For the on Correlations option, the 1st principal component is a linear combination of the centered and scaled observations using the entries of the 1st eigenvector as coefficients.

But what I understrand from how eigengenes were originally calculated, SVD was used.

TRR21

Community Trekker

Joined:

Nov 4, 2016

Re: Eigengene values from a set of gene expression values

Genes as columns. Genes to be used haven't been designated in any particular way - was thinking I would specify during the GUI dialog or in the script in order to generate the eigengene/or equivalent for each of the samples (rows). Thanks.