JMP Genomics for plant breeding analyses and prediction
Aug 30, 2016 5:10 AM(1904 views)
I have been watching Russ Wolfinger's presentation on breeding analyses and prediction. Alas, I am not able to copy him on my own computer. I suspect that I miss some important stuff. For instance, I cannot copy how he sets up the 4-5 ridge regression analyses.
Are there written tutorials for the breeding stuff? I'd be happy to see that.
First, find the "Cross Validation and Model Comparison" process in the JMP Genomics menu. The menu path to this process is: "Predictive Modeling >> Model Comparisons >> Cross Validation and Model Comparison". Once you reach this process, click on it to open the process window. In the new window, go to the "General" tab, where you will choose the folder that holds the setting files for the models you want to compare by clicking on "Choose" button for the "Folder of Setting Files" field. Choose the folder path where the setting files are located. If you used the "Predictive Model Review" process to create the setting files, then choose the output folder for the "Predictive Model Review" as the "Folder of Setting Files". At this point, you will see on the left panel all available setting files. Click and highlight all settings you want to run the cross validation and model comparison on and click on the arrow "-->" sign to send them to the right panel. You can set the output folder too in the General tab. Now, go the "Analysis" tab and choose the "Hold-Out Method" (a good one is the Random Partition), also choose the "Hold-Out Size, Specify as:" ("K for K-Fold or 1/K Hold-Out" is pretty standard, so you can use it). Next, set a number for "K for K-Fold or 1/K Hold-Out" (if you have sample size of 100 and choose K=4, then the data set will be split into four parts, in turn each one of them will be used as training data set, and the rest as test set). Finally, set a number for "Number of Random Hold-Out Iterations" (this the number of times the cross validation will be processed; the greater the better, however, it can be time consuming, so I would suggest trying out 5). You may leave all other options as default.