I try to annotate one new genome, several sources of evidence were used, include ESTs (PASA), homolog protein aligments (Genewise) and abinito predictions,

The problem is that i got several gene models for each locus, for example, GeneWise alignment create >500,000 gene models, I run wise after blast, but I do not filter the blast result (all blast hits were used for genewise alignment).

I guess that blast results should be filtered before running wise alignment, but I don't know routine paramers that can be used to filter blast results (eg. identity, alignment length)

dose anybody have some opinion about the routine process in gene structure prediction?