Is there a protocol for "adjust metarpars" after WebAUGUSTUS has problems with accuracy?

[SchwarzEM]

I have used WebAUGUSTUS to generate AUGUSTUS species-specific parameters for a parasitic nematode genome with high accuracy (1e-04 error rate) and contiguity (chromosomal single scaffolds). Two other genomes of this quality were very easy to get decently high accuracy for AUGUSTUS (>= 0.80) using the BRAKER2 pipeline, but this particular genome for unclear reasons had repeatedly bad accuracy and gene predictions in BRAKER2, so I instead tried training AUGUSTUS with either cDNA or protein via WebAUGUSTUS. My problems with this particular genome continued: the accuracy of protein-based AUGUSTUS parameters was mediocre (around 0.50) but not horrible; for cDNA-based parameters, the accuracy was quite awful (around 0.18!).

In both cases, I got the following error messages:

0 WARNING: optimization did not improve accuracy!
1 WARNING: Optimizing didn't improve the accuracy, maybe need to adjust XXX_metarpars!

Which I would be happy to do, except ... how? How exactly do I go about adjusting the meta parameters? Is there a protocol for doing this? I would very much like to find one.

[KatharinaHoff]

The species (let's call it crazy_worm) has a file crazy_worm_metapars.cfg in the Augustus config/species/crazy_worm folder. This file contains the value ranges that are used during optimize_augustus.pl. It is possible that the default parameters range does not work well for a particular species, and you manually adjust the values listed there. Then run the training, again (optimize_augustus.pl). Check accuracy. Maybe it helps.

I would advise to first look at the gene structures in a browser and try to figure out what exactly is the issue. That makes it easier to identify a candidate parameter that needs adjustment.

One issue that I have been fighting with in several genomes is the poly-signal. The parameters for this are in the crazy_worm_parameters.cfg file at the bottom. You can manually adjust them if you know e.g. that this worm has a different poly-a consensus.