Lassosum data input requirements

[lambdr]

In the original paper describing lassosum, the package was developed for the scenario that you have the summary statistics and a set of PLINK files. It does not take an LD matrix as input, and the PLINK reading is built into the C++ component of the pipeline.

The method never calculates the whole LD matrix explicitly but considers each variant iteratively in the original C++ code. The approach used is a coordinate descent algorithm to estimate the beta coefficients for LASSO. Here they update beta for variant i in iteration t.

Our current understanding is if you can expand equation 11, the X-tilde terms will reduce to an element/row of the LD matrix. And therefore, it is possible to apply this algorithm to perform LASSO on summary statistics, but that would require development of a new set of C++ functions.

[gaow]

Thanks @lambdr . @danielnachun see above -- it is not straightforward to include lassosum in a way compatible to our current interface of summary stats and LD matrices.