L2-penalty LD shrinkage

[xiangzhu]

I wonder how hard to implement the following L2-penalty (i.e. "ridge") LD shrinkage estimator: Sigma_ridge = Sigma + lambda I, where Sigma is the sample covariance matrix for SNPs.

For simplicity we can fix the value of lambda (e.g. lambda=0.01) for now. (It might be conceptually straightforward to tune this parameter by cross-validation, but this definitely complicates the software implementation.)

This approach was used in ImpG-Summary: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4184260/