Gravar-mail: Using reads to annotate the genome: influence of length, background distribution, and sequence errors on prediction capacity