I run SNAP in MAKER using RepeatMasker to produce initial models, so it will
run on masked sequence by default. You can run on non-masked sequence as
well if you really want to by setting unmask in the control files, or by
turning off all repeat masking options.
The details on how masking is handled in other training steps (like
maker2zff, forge, etc.) is a little more complicated, but is automatically
handled for you correctly. So you don't need to worry about it there.