cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
  • JMP 19 is here! See the new features at jmp.com/new.
  • Register to attend Discovery Summit 2025 Online: Early Users Edition, Sept. 24-25.
Choose Language Hide Translation Bar
qunw
Level III

Test and Build ONE to Many Correlations

Hi JMP community, 

 

Does any of you know how to test One-to-Many correlations instead of using the "multivariate" function at Aanalyze?  I would like to quickly screen and generate a top list of genes from a 20,000-gene panel that have strong correlations with the target of interest, along with corresponding r and p-values. Then, build the correlation matrix with the target of interest.  Many thanks!

6 REPLIES 6
Victor_G
Super User

Re: Test and Build ONE to Many Correlations

Hi @qunw,

I think the platform Response Screening might be helpful for you.


In your case, this platform enable you to test all your X with your response of interest Y, with the use of appropriate robust test to control false discovery rate (due to the large number of tests conducted).
Volcano plots (Logworth by Difference and Logworth by Slope) are also available in this platform, and are commonly use in genomics to identify the most impactful genes on a response.

Hope this answer will help you,

Victor GUILLER

"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
qunw
Level III

Re: Test and Build ONE to Many Correlations

Thanks, Victor, for your instruction!  Yes, Response Screening is useful for identifying statistically significant X that affects the response of interest Y.  My query above aims to identify the most correlated X with a target of interest, which is also a marker within the X group.  I know that multivariate allows building a correlation n*n matrix, but I would like to see one *n correlation in X group, with corresponding r and p-values.  Any suggestions on this?  Many thanks!

Victor_G
Super User

Re: Test and Build ONE to Many Correlations

Ok, not sure I fully get your response, but maybe Cluster Variables platform might be helpful ?
You'll have the Color map on Correlations results and the Cluster summary will enable to group your variables into clusters, providing cluster component variables (principal component by cluster).
However, no p-values displayed in this platform, this information can be found in Multivariate platform.
If this response still does not answer your question, could you precise your objective, with a toy dataset, screenshots, results expected, ... ?

Hope this may help you,

Victor GUILLER

"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
qunw
Level III

Re: Test and Build ONE to Many Correlations

Hi Victor, 

I appreciate your help with the query.  Here's the example of one-to-many correlation output:  Do you know how to do it with JMP?

Many thanks!

qunw_0-1753836792204.png

 

Victor_G
Super User

Re: Test and Build ONE to Many Correlations

Ok, this is typically the type of output you'll get with the platform Response Screening when adding one response and all the factors in the Launch dialog, with some additional options (different plots) and with more protection against multiple comparison false discovery rate risk thanks to Benjamini-Hochberg technique for calculating FDR p-value, and additional robust calculation options.

 

If this is not what you want, I don't understand what's missing or what's not complete for your needs. The platform Fit Y by X also enables to do bivariate analysis, but in your case regarding the number of variables to test, the platform may not be adapted, as it will be cumbersome and statistical tests won't be adapted regarding the number of comparisons done.

 

Note that you can always extract datatables in reports (right click in any table report and choose "Save as Datatable" to get the JMP Datatable), if you want to customize plots with the Infos extracted from a platform ; for example, extracting infos from the Response Screening report to plot your own scatter plot matrix for visualizing correlations with p-values.

 

Sorry if this response doesn't answer your needs, I'm out of options,

Victor GUILLER

"It is not unusual for a well-designed experiment to analyze itself" (Box, Hunter and Hunter)
qunw
Level III

Re: Test and Build ONE to Many Correlations

Thanks so much, Victor, for all your suggestions on this query!  I also reached out to JMP support and learned that they don't have an exact feature in JMP 18 for One-to-Many, but will add it to the Wishlist.  For now, I have it done by Spotfire; in case you have similar needs, that's complementary to JMP for this function until the new feature is added.  

Recommended Articles