Combine BRAKER3 and Galba annotation for new, non-model genome

[YanisChrys]

Hello,

In order to better annotate my reptile genome I have used BRAKER3-ETP and Galba (with ~6 of the closest relative proteomes) and now I am trying to combine them using TSEBRA. The inputs look like this:

BRAKER3 | C:67.6%[S:56.0%,D:11.6%],F:3.6%,M:28.8%,n:7480

GALBA | C:93.3%[S:75.1%,D:18.2%],F:1.6%,M:5.1%,n:7480

This is the command I am using:

singularity exec --bind $WORKDIR /path/to/containers/braker3.sif /opt/TSEBRA/bin/tsebra.py \ --gtf ${galba_gtf},${braker_gtf} \ --hintfiles ${hints_gff2},${hints_gff1} \ --filter_single_exon_genes \ --ignore_tx_phase \ --cfg ${braker_cfg} \ --out ${out_gtf} \ --verbose 1 2> ${log}

I have experimented a lot but I can't find a way to run it that maximises busco and keeps mostly single genes.
I have tried the following:

• Adding ${augustus_gtf},${genemark_gtf} from BRAKER3 run
• Using different config files and thresolds:
  • Braker3.cfg, default.cfg, pref_braker1.cfg
  • custom cfg with:
  • e_1 =0.1, e_2-4=0.05
  • e_1-4=0.05 | 0.005 |
  • other combinations where either e_1 is higher or lower than the other 3
  • increasing or decreasing intron_support
  • increasing or decreasing "P" and "E" weights

However these all result in a lot of duplicates (65-87%) and only slightly better busco than galba (92-94%).

Decreasing the e_n values, reduces duplicates but also the busco score.

Perhaps the issue is with transcripts having equally low support so more than 1 overlapping transcripts are kept?

Any advice is welcome. Please feel free to ask for any extra information or log files.