When imputation is involved, I was initially worried that it is no longer reasonable to store data as sparse matrix because those homozygous reference alleles will be re-imputed and assigned non-zero values (of allele dosages) for most of them. At least this is what I've seen in my data in the past. However I recently used UMich's imputation server for my project. I found that the server does leave the originally deemed homozygous reference genotypes as is. There does not seem an option to configure it otherwise. I think this imputatoin server is almost the gold standard these days. In that case we may get away with relying on sparse matrix not worrying about imputed data.
When imputation is involved, I was initially worried that it is no longer reasonable to store data as sparse matrix because those homozygous reference alleles will be re-imputed and assigned non-zero values (of allele dosages) for most of them. At least this is what I've seen in my data in the past. However I recently used UMich's imputation server for my project. I found that the server does leave the originally deemed homozygous reference genotypes as is. There does not seem an option to configure it otherwise. I think this imputatoin server is almost the gold standard these days. In that case we may get away with relying on sparse matrix not worrying about imputed data.