Genetic Association with JMP Genomics, Part 3c: Q-K Mixed Model

[#1]

Q-K association analysis was developed to perform association mapping while controlling for population structure and/or familial relatedness. The “Q-K” in this name refers to the two kinds of information that get included in the model. The Q matrix contains information about population structure (See Part 3b: Population Structure Matrix). The K matrix contains more fine-grained information about relatedness, usually IBD measures calculated from the marker data (See Part 3a: Marker Based Relationship Matrix).

As in other kinds of association mapping, in Q-K association analysis an individual statistical model is created for each marker, using the trait as a dependent variable and the marker as an independent variable. The variables that constitute the Q and K matrices are also included in these models: Q variables as fixed effects, and K variables as random effects. Either the Q or the K variables may be omitted from the models, as desired.

To run Q-K association analysis in JMP Genomics, three things are needed: a data file containing the traits to be analyzed, the marker genotypes, and the Q and K variables to be used in the analysis. An annotation data set is optional.

1. Use the File > Open command to open the PCA results merged file, called pca_output_pcm.sas7bdat
2. Also Open the file with the “square root” of the IBD matrix from the Relationship Matrix procedure, called rice_genos_recgeno_rm.sas7bdat.
   • Note that this file should be replaced with the compressed K matrix file, rice_genos_recgeno_ibd_kc.sas7bdat, if applicable.
3. With the file rice_genos_recgeno_ibd_kc.sas7bdat in the foreground, select Tables > Join.
4. In the top left box, select the file pca_output_pcm.sas7bdat
5. Check the box labeled Merge same name columns
6. Select the variable GID in the two Source Columns boxes, and then click the Match
7. Assign the Output table name as core_qk.
8. Click OK to create the new joined file.
9. When the new data file appears, select File > Save As…
10. Change the Save as type option to SAS Data Set and click Save
11. From the Genomics Starter menu, choose Genetics > Other Association  Testing > Q-K Mixed Model.
12. Choose core_qk.sas7bdat as the Input SAS Data Set.
13. Select GW as the Trait Variable.
14. Enter recgeno: in the box under List-Style Specification of SNP Variables.
15. Choose an Output Folder.
16. On the Q and K tab, type pca: in the List-Style Specification of Q Matrix Variables box, and type IBD: in the List-Style Specification of K Matrix Square Root Variables
    • If using the compressed K matrix, check the box K Matrix is compressed.

17. On the Model Variables tab, select Continuous for Type of Trait. There are no additional effects in this data set.
18. The Model Selection tab allows for the implementation of a multi-marker model, based on rotated markers and phenotypic variables. When selected, the user can choose from methods like ElassticNet, Stepwise, and LASSO. We will not fit a multi-marker model in this example.
19. On the Annotation tab, Choose rice_anno_recgeno.sas7bdat as the Annotation SAS Data Set.
20. Fill out the Annotation tab with RS_RG as the Annotation Label Variable, chrom as the Annotation Group Variable, pos as the Annotation Location Variable, and _MajorAllele_ as the Annotation Major Allele Variable.
21. On the Options tab, change the Format of SNP Variables to Numeric Genotypes.
22. For a shorter computing time, deselect the Genotype Association Test. Optionally, the Genotype test performs chi-square tests based on genotypes, where the Trend test performs a Cochran-Armitage test.
23. Click Run to start the analysis.

Results

24. When the results dashboard opens, the Summary Chart tab shows the number of significant markers on each chromosome for the Grain Yield (GW)
    • The blue bars represent results from the Trend test. The trend test looks for a linear relationship in the trend scores when moving from homozygous minor to heterozygous to homozygous major.
25. When performing a model on multiple traits, significant markers that overlap across the traits can be viewed and data subsets can be created via a venn diagram output.
    • There are five markers that are significant in common between GW, FL, and PH.
26. The Manhattan Plot tab once again shows significant markers by chromosome as determined by the trend test.
27. Selecting any of the chromosomes from the Tabs menu on the left will bring up a plot of each marker on the chromosome plotted by position. Shown below is chrom 3 Results.
28. The Model Fitness Plot tab shows fitness statistics for each marker. To separate the fitness plots by model, click the red triangle and select Overlay Plots > Overlay Ys. Tests are plotted by shape, with points lower on the plot representing higher accuracy.
29. The Volcano Plot tab gives a volcano plot for each marker, with minor allele genotype effect on the x-axis and the log transformed p-value on the y-axis. Points above the red line are considered significant.
    • Clicking through the Local Data Filter options on the right will give volcano plots for each individual chromosome.
30. In the Drill Downs menu, a subset can be created for selected markers, or a plot of Grain Yield values by genotype can be created by selecting Plot Trait By Genotype.
    • This is a great way to quickly see the effect different genotypes have on a trait for a given marker.
    • Shown below are trait by genotype plots, ordered by significance, from three markers on chromosome 5.

*The interactive results from this analysis are available on JMP Public.

Summary

This guide outlined the the Q-K Mixed Model process for association mapping while accounting for linkage disequilibrium caused by population structure and familial relatedness. QK analysis can be run in two different ways in JMP Genomics. A simpler option is the Genetics Q-K Analysis Workflow, which performs all the steps for the Q-K analysis and merges the data automatically. It is, however, less flexible than the process outlined in this post. The workflow only allows a Q matrix computed from PCA, and a K matrix from IBD calculations from the Relationship Matrix process. For detailed information on this process, see next week’s blog post: Part 3d: Q-K Mixed Model Genetic Association Workflow.

rice_genos_recgeno_ibd_kc_anno.zip