Evidence codes for GO predictions by neural network tools

[PSB-mibel]

Hi,

The IEA evidence code seems to be a left-over bag to hold any automatic electronic predictions.
However, I feel there is a difference in confidence between the annotations produced by InterPro-scan (with the help of interpro2go) and the results produced by the wide variety of newer GO prediction tools (e.g. https://github.com/flatironinstitute/DeepFRI). Especially given that quite some effort has gone into the manual curation of the interpro2go mapping in itself, and the (relative) very low false positive rate of InterPro-scan in itself.

As such I think it would be advantageous to discuss whether introducing more diverse evidence codes (rather than simply labeling all these annotations as IEA) would be a good approach to avoid the issue where the same confidence is attached to all methodologies.