rmscca

robust multiple sparce canonical correlation analysis

As described in the manuscript, Robust Sparse Multiple Canonical Correlation by J. Coleman, J. Replogle, G. Chandler and J. Hardin. Available at http://arxiv.org/abs/1410.3355

The functions needed in the analysis are the following:

All programs

• scca.CVperm (does the cross validation and permutation -- all the work to get over curse of dimensionality)
• scca.function (does the thresholding for calculating lambda values. Very important function for scca.CVperm)
• sample.sigma12.function (calculates the cross covariance matrix, K)

Only simulating (both null and correlated data)

• Cov.suped (makes the correlation matrices for X and Y)
• sim.setup (simulates the data!)
• build.B (generates the relationships between X and Y)

Interpreting results

• parse.breast (count things from the breast cancer results)
• interpret.results.curveonly (uses the Q-permutation curve to count positives / negatives)
• results (more parsing output)
• results.helper (more parsing output)
• determine.true.vals (for the blocking)

To run

Null: nullSimScript.R

- generates null data with k=0 (i.e., B==0).  Only uses the first CC to determine if anything is called significant.
- type I error is calculated using null_results.R

Breast cancer data: dataScript2.R (only chrom2) dataScript.R (all chrom in parallel)

- Uses the .csv files from the PMA package.  
- the main function doing the work is scca.CVperm
- the output contains the list of coefficients and correlations for both spearman and pearson

Simulated data: fullSimScript.R

- Simulates data, runs the RMSCCA code, counts things like "complete groups", false positives, etc.  Output is as given by interpret_results_curveonly.R.  
- Data is parsed and plotted using plot_results.R.