GAF file format specs

[ialarmedalien]

Is there a machine-readable version of the GAF file format specification somewhere? I have found the GPAD and GPI specs but haven't found a similar file for the GAF format. Any format is fine - linkml, JSONschema, etc.

I know I can reconstitute the GAF xSV format from the GPAD/GPI specs.

[kltm]

@ialarmedalien Yeah, this is not the best--the GPAD/GPI 2.0 spec is very nicely done. I think the most honest answer is that we've usually just depended on the various parser implementations, without writing a nice formal grammar-like spec like that. There is this pretty-good-for-many-purposes GAF 2.2 spec (https://geneontology.org/docs/go-annotation-file-gaf-format-2.2/), and while there has been a pass at a LinkML model for GAF, I don't believe we have one that is really supported.

That said, depending on how you're approaching this and how you're defining machine-readable, we also have the GORULES (https://github.com/geneontology/go-site/tree/master/metadata/rules), which goes into pretty good detail about what would be considered legal or illegal in GAF. (Our soon-to-be-legacy ontobio is built up and tested against this).

[ialarmedalien]

What is ontobio going to be replaced by? I was considering using that to parse GAF files, but if it's going to be deprecated, I'd sooner use something with a bit more shelf life.

Additionally, is there a utility somewhere to generate GPAD and GPI files from a GAF file?

[kltm]

@ialarmedalien TBD, but it will be python and built around LinkML.

I can probably find our experiment with LinkML and GAF somewhere, but I believe we'll be doing it clean when we turn back to it. For the moment, with more of the parsing and QC of the tabular files being done upstream at GOA and our focus on GO-CAM, the main off-the-shelf tool I'd reach for would be ontobio, in spite of it being deprecated. We have our GO-CAM LinkML fairly well locked in, and we should be making our way back down the file parsing back to GPAD/GPI and GAF, but it's not in the roadmap yet. (That said, that could change quickly and we'd likely make rapid progress.)

To answer the second question, ontobio can handily do the file conversions you'd want as well, but it's not something we'll be putting much more direct effort into.

If you want something now that does all of the above, ontobio would be what I point to.
If you are looking to wait, we'll (at some point) produce something python and LinkML.
If you are going to forge ahead, we'd recommend python and LinkML (and would love to give feedback or fold back in over here if it works out).