Evidence Legend

[JacobsonMT]

Need a legend for evidence types.

Ex. Automatic (IEA), etc...

[ppavlidis]

I believe I had a comment somewhere else that this was kind of wonky. The main distinction people make is IEA vs. everything else. Right now it is not clear where IEA lives.

The current "Evidence Breakdown" provides a link to the exlanation of evidence codes from GO, but the image doesn't expand on click (it's https://gotrack.msl.ubc.ca/javax.faces.resource/diag-evCodeFlowChartGood.png?ln=img)

And our categories don't match that. They have "experimental" "non-experimental" "author statement" and "curator statements". We have five categories.

I wonder if, for the web interface, we shouldn't pool all "statements" (which are a minority) and then "experimental", "non-reviewed computational" and then "IEA"?

[JacobsonMT]

The flowchart is out of date. We use the categories shown on the GO website (http://www.geneontology.org/page/guide-go-evidence-codes) specifically:

• Experimental
• Computational
• Author
• Curator
• Automatic

Should maybe just replace the flowchart with a link to this page?

[ppavlidis]

My original idea was to simply pool Experimental, computational, author, and curator into one group ("curated"), and "automatic" is just the IEA code. A lot of biologists (and bioinformaticians in particular) know that distinction. I felt the categories are opaque to most users of GO/A while the IEA vs "other" split is better known. In particular "computational" and "automatic" don't sound any different from the name.

However, it sounds like GO is trying to eliminate the IEA code (they say "GO also used one evidence code ..." - since when is it past tense?). And it's interesting that now GO says "Annotations using the IEA code should be reviewed after one year, any older than this date will be deleted". Is this actually true? It's news to me, assuming "reviewed" implies "turned into another evidence code".

So perhaps it would just help to rename the five categories to something more like:

• Curated experimental
• Curated computational
• Curated author statement
• Curated inferred
• Non-curated (IEA) --- and I mean put "(IEA)" as part of the label since it's pretty well known.

As well as linking to that page from GO.

The underlying question here is: do these categories say something about how accurate the annotations are? Unsubstantiated lore is that non-curated annotations are worse (an evaluation of this by one group suggested the lore was wrong, or at least hard to substantiate). But among the other evidence codes I am not aware of anybody caring (much less looking).