High rate of non-canonical splice site predictions

[EleanorEveCarr]

Hello, thank you for creating Helixer.

We've discovered an issue when running Helixer on Boeremia exigua RefSeq genome assembly GCF_020726555.1 https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_020726555.1/.
Helixer is predicting non-canonical splice sites for ~31% of genes, while the original RefSeq annotation for this same assembly contains only 0.9% of genes with non-canonical splice sites (which is within the normal range for fungi).

• Helixer version: v0.3.3
• Model: fungi_v0.3_a_0100.h5

We also noticed that this assembly was included in Helixer's training data for model fungi_v0.3_a_0100.h5.

Here is an example gene where Helixer incorrectly predicts non-canonical splice sites for both introns despite the RefSeq annotation showing canonical sites very close by:

Gene: gene-C7974DRAFT_127213
RefSeq GFF (canonical):

NW_025763520.1  RefSeq  gene    1287542 1288678 .       +       .       ID=gene-C7974DRAFT_127213;Dbxref=GeneID:70164704;Name=C7974DRAFT_127213;gbkey=Gene;gene_biotype=protein_coding;locus_tag=C7974DRAFT_127213
NW_025763520.1  RefSeq  mRNA    1287542 1288678 .       +       .       ID=rna-XM_046135284.1;Parent=gene-C7974DRAFT_127213;Dbxref=GeneID:70164704,GenBank:XM_046135284.1;Name=XM_046135284.1;gbkey=mRNA;locus_tag=C7974DRAFT_127213;orig_protein_id=gnl|WGS:JAHBNH|C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;transcript_id=XM_046135284.1
NW_025763520.1  RefSeq  exon    1287542 1287905 .       +       .       ID=exon-XM_046135284.1-1;Parent=rna-XM_046135284.1;Dbxref=GeneID:70164704,GenBank:XM_046135284.1;gbkey=mRNA;locus_tag=C7974DRAFT_127213;orig_protein_id=gnl|WGS:JAHBNH|C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;transcript_id=XM_046135284.1
NW_025763520.1  RefSeq  exon    1287954 1288365 .       +       .       ID=exon-XM_046135284.1-2;Parent=rna-XM_046135284.1;Dbxref=GeneID:70164704,GenBank:XM_046135284.1;gbkey=mRNA;locus_tag=C7974DRAFT_127213;orig_protein_id=gnl|WGS:JAHBNH|C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;transcript_id=XM_046135284.1
NW_025763520.1  RefSeq  exon    1288414 1288678 .       +       .       ID=exon-XM_046135284.1-3;Parent=rna-XM_046135284.1;Dbxref=GeneID:70164704,GenBank:XM_046135284.1;gbkey=mRNA;locus_tag=C7974DRAFT_127213;orig_protein_id=gnl|WGS:JAHBNH|C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;transcript_id=XM_046135284.1
NW_025763520.1  RefSeq  CDS     1287670 1287905 .       +       0       ID=cds-XP_046000208.1;Parent=rna-XM_046135284.1;Dbxref=InterPro:IPR008701,JGIDB:Boeex1_127213,GeneID:70164704,GenBank:XP_046000208.1;Name=XP_046000208.1;gbkey=CDS;locus_tag=C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;protein_id=XP_046000208.1
NW_025763520.1  RefSeq  CDS     1287954 1288365 .       +       1       ID=cds-XP_046000208.1;Parent=rna-XM_046135284.1;Dbxref=InterPro:IPR008701,JGIDB:Boeex1_127213,GeneID:70164704,GenBank:XP_046000208.1;Name=XP_046000208.1;gbkey=CDS;locus_tag=C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;protein_id=XP_046000208.1
NW_025763520.1  RefSeq  CDS     1288414 1288485 .       +       0       ID=cds-XP_046000208.1;Parent=rna-XM_046135284.1;Dbxref=InterPro:IPR008701,JGIDB:Boeex1_127213,GeneID:70164704,GenBank:XP_046000208.1;Name=XP_046000208.1;gbkey=CDS;locus_tag=C7974DRAFT_127213;orig_transcript_id=gnl|WGS:JAHBNH|C7974DRAFT_mRNA127213;product=necrosis inducing protein-domain-containing protein;protein_id=XP_046000208.1

Helixer prediction (non-canonical):

NW_025763520.1  Helixer gene    1287446 1288663 .       +       .       ID=_NW_025763520.1_000229
NW_025763520.1  Helixer mRNA    1287446 1288663 .       +       .       ID=_NW_025763520.1_000229.1;Parent=_NW_025763520.1_000229
NW_025763520.1  Helixer exon    1287446 1287905 .       +       .       ID=_NW_025763520.1_000229.1.exon.1;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer five_prime_UTR  1287446 1287669 .       +       .       ID=_NW_025763520.1_000229.1.five_prime_UTR.1;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer CDS     1287670 1287905 .       +       0       ID=_NW_025763520.1_000229.1.CDS.1;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer exon    1287957 1288365 .       +       .       ID=_NW_025763520.1_000229.1.exon.2;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer CDS     1287957 1288365 .       +       1       ID=_NW_025763520.1_000229.1.CDS.2;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer exon    1288417 1288663 .       +       .       ID=_NW_025763520.1_000229.1.exon.3;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer CDS     1288417 1288485 .       +       0       ID=_NW_025763520.1_000229.1.CDS.3;Parent=_NW_025763520.1_000229.1
NW_025763520.1  Helixer three_prime_UTR 1288486 1288663 .       +       .       ID=_NW_025763520.1_000229.1.three_prime_UTR.1;Parent=_NW_025763520.1_000229.1

Are there any recommended parameters to use for cases like this? Any suggestions for troubleshooting this would be greatly appreciated!

Thanks in advance.

[felicitas215]

Hi, thank you for your question.
We have looked into it, and you're right. Helixer has extended both introns by 3 base pairs (when compared to the reference) and thereby ended up with different 3' intron splice sites in both cases.

Feature	start	end	length (bp)	annotation_source
Intron 1	1,287,906 (GT)	1,287,956 (GG)	51	Helixer
Intron 1	1,287,906 (GT)	1,287,953 (AG)	48	Refseq
Intron 2	1,288,366 (GT)	1,288,416 (AC)	51	Helixer
Intron 2	1,288,366 (GT)	1,288,413 (AG)	48	RefSeq

We have a constraint in HelixerPost that for U2 splicing the intron needs to be at least 49 bp long. This was done based on literature which suggests the minimum U2 intron length of 50 bp, but it seems that there are quite a few exceptions to this (Fungi are in this category). We will investigate how to we can adapt this to different species, so we will mark this issue as an enhancement.

[EleanorEveCarr]

Thank you for looking into this. Just to note, we were able to confirm this pattern for this annotation: we extracted all of the transcripts with zero canonical splice sites and calculated the lengths of their introns - the majority fall within 50–52 bp, which is in line with this constraint.

[BjoernUsadel]

Depending on how you feel about it and if you can do three step basecalling.
You can try https://github.com/BjoernUsadel/HelixerPost. As I can't currently basecall while working on Intel iGPU support in Helixer, I couldn't test it with your data.
Please note that this is not a drop in replacement as it uses named parameters now ./target/release/helixer_post_bin genome_data.h5 predictions.h5 wurst2.gff -w 100 -e 0.1 -p 0.8 -m 60 --min-intron-len-u2-gt-ag 45 --min-intron-len-u2-gc-ag 45 --min-intron-len-u12-at-ac 29. This is lowering intron length to 45,45,29. Normal values are 49,49,29.