search

yuanzhongshang / PMR

22 stars 7 forks source link

Dealing with singular matrix #6

Open seokho92 opened 1 month ago

seokho92 commented 1 month ago

Hi, I am trying to use PMR on imputed pQTL cis region, with 100kb up/downstream.

This usually leaves 500~1500 markers on each gene, and many times it goes to a pitfall in cholesky decomposition because of singularity.

These are three strategies I tried so far:

  1. Pruning based on pairwise correlation / VIF
  2. Increase diagonal terms to ensure positive definiteness
  3. Increase lambda (In worst cases, increasing lambda does not fail only when it is given as > 5.)

Any feedbacks, suggestions will be very helpful. Thanks!

yuanzhongshang commented 1 month ago

Hi,

Thanks for the interest with PMR paper. The pruning strategy should be preferred, but you can try pruning and diagonal terms increase to compare and check the robustness of the results.

Best, Zhongshang