Say I have a gentoype matrix where I use the encoding

0 = HOM_REF
1 = HET
2 = HOM_ALT
NA = Missing genotype

For instance this dummy genotype matrix with 3 variants and 3 samples

Variant_1    0 1 1 
Variant_2    1 1 NA
Variant_3    NA 0 1 
etc

Do you need to first impute the genotype matrix to not have any missing genotypes(NA values)?

Or do you set the NA values to something like -9 or -999? This would influence the output of the linear / logistic regression heavily for variants with a lot of missing genotypes?