bg7 is an annotation system where the ORF prediction step is based on protein similarity. That’s the key feature, and follows what they view as the guiding principle: use as much information as possible:
Homepage: http://bg7.ohnosequences.com/
Instead of relying almost exclusively on start and stop signals (as some other pipelines do for ORF prediction) bg7 relies on the overall similarity of the region to a certain protein. It then looks for start/stop signals to define the ORF. If it does not find them (suppose they’re pseudogenes or there are lots of sequencing errors) it will mark it as putative ORF.
See also the slides on slideshare