gene-annotation-refiner mode 2 dramatically reduce busco C and introduce‌ premature termination genes

[xiekunwhy]

Hi,

I am trying to use gene-annotation-refiner to refine gene models generated from evm(https://github.com/EVidenceModeler/EVidenceModeler) pipeline with following command line,

python gene-annotation-refiner/gene_annotation_refiner_2.py --genome rbu.fa --output refined_v2.gff --refine_existing EVM.all.gff --bigwig out.bw --junctions rbu.junction.merge.bed --refine_with_evidence

after this step, I use gffread to get protein sequences with or without -V -J options

gffread -g rbu.fa -y refined_v2.all.pep.fa refined_v2.gff
grep ">" refined_v2.all.pep.fa|wc -l
32386

gffread -g rbu.fa -y refined_v2.J.pep.fa -J refined_v2.gff
grep ">" refined_v2.J.pep.fa|wc -l
31937

gffread -g rbu.fa -y refined_v2.V.pep.fa -V refined_v2.gff
grep ">" refined_v2.V.pep.fa|wc -l
32368

gffread -g rbu.fa -y EVM.all.pep.fa -J EVM.all.gff
grep ">" EVM.all.pep.fa|wc -l
32727

There are 32727 genes in EVM.all.gff, 32386 left (or produced) after refining, but there are 32368-31937=431 gene are premature termination.

Here are busco evaluation (embryophyta_odb10) results using EVM.all.pep.fa, refined_v2.V.pep.fa and refined_v2.all.pep.fa as input respectively,

EVM.all.pep.fa as input
C:97.5%[S:87.1%,D:10.3%],F:1.4%,M:1.2%,n:1614

refined_v2.V.pep.fa as input
C:92.7%[S:83.8%,D:8.9%],F:4.5%,M:2.8%,n:1614

refined_v2.J.pep.fa as input
C:92.7%[S:83.8%,D:8.9%],F:4.5%,M:2.8%,n:1614

Busco C value is reduce dramatically, so the result may not so reliable using RNA evidence only to refine consensus genes from up stream pipelines ?

Best,
Kun