Hi Hail Team,

I’m using realized_relationship_matrix() to generate a pre-computed kinship matrix for use with linear_mixed_model() and have found that it’s quite slow. Under the hood, I see that it is only using a single process, which I imagine is a factor. Do you have any suggestions for improving performance? I assume more stringent LD pruning before running it would mean fewer variants going into the function, and would improve performance? Is this something you’d suggest doing? Are there any guidelines for the number of variants that is optimal to use? Finally, is it somehow faster if the genotypes are provided directly to linear_mixed_model() with the z_t argument vs pre-computing? Sorry for the flood of questions, and thanks in advance for any help.

Best,

Stephane