There is a small utility for reporting feature overlaps between gff files.

It requires a bit of editing (sed-ing rather) of typical gff files:

gff file: it should contain at least 8 fields separated by tabs, followed by a list of attributes of the form [key space value] separated by spaces. There should not be any space at the end

sed 's/[;=]/ /g' filename.gff

does the job. I just started testing it (works on my test files) but did not figured out complete usage.

Yet another gene prediction comparison software, Eval from Brent lab:

http://mblab.wustl.edu/software/eval/

Works OK on provided .gtf example files, giving a very detailed output. Gets confused by gff3 files from Augustus and SNAP, does not seem to understand exonerate output reformatted to gff3. Takes a long time to load Augustus 1.4M lines gff file.

I believe the answer you are looking for is ParsEval, which is now part of AEGeAn software toolkit. This is by some of the same folks who made GAEVAL.

• Daniel S. Standage and Volker P. Brendel (2012) ParsEval: parallel comparison and analysis of gene structure annotations. BMC Bioinformatics, 13:187, doi:10.1186/1471-2105-13-187.
• http://parseval.sourceforge.net/
• http://aegean.readthedocs.org/en/latest/

There is some rather old Java program for comparing prediction accuracy: http://bioinformatics.oxfordjournals.org/content/19/13/1712.abstract

ftp://iubio.bio.indiana.edu/molbio/genefind/

I had no time to test it yet.

The developers of PlantGDB developed a GAEVAL tool (Gene Annotation Evaluation), but this basically scores gene models based on aligned evidence.

I am currently running a genome annotation pipeline and I would like to do the same thing--compare the predictions from the annotation pipeline with the annotations provided by the group that did the sequencing and assembly. I wasn't really interested in Jigsaw, since it takes the two (possibly) disparate gene models and builds a consensus gene model. I'm more interested in what you mentioned--summary statistics about the differences between the two sets of annotations.

I am currently working on a script to do this comparison. I'm still trying to figure out what comparisons/statistics are interesting (borders of gene model, exon agreement, etc). Will share when I have something reliable.

There is JIGSAW, which reads GFF. It does somewhat more than you're asking and combines the input into new gene models.

I didn't understand very well about the mode flag . Could you please give some examples? I am using some sed commands in order to make some conclusion like how many genes structures are identical between two sets. Is that also possible by any flag.