[BVBRC] pseudogene/partial_cds annotation gap for genomes

[jananiravi]

Problem: pseudogenes/invalid genes stall downstream analysis with BV-BRC genomes

Summary

Genome 1282.2719 contains at least 25 CDS features whose translated sequence has an internal premature stop codon — i.e., verifiably non-functional/pseudogenized ORFs — none of which are reflected in any BV-BRC-reported metadata field. The GFF's own pseudogene feature-type annotation reports 0 for this genome, and the genome-summary partial_cds/partial_cds_ratio fields also report 0. This genome additionally lacks ##sequence-region header lines in its GFF3 (present in every other genome we checked) and uses self-referential placeholder contig IDs (see below), suggesting it was processed through an older or different annotation pipeline than typical current BV-BRC submissions.

Evidence

Method summary (the "how")

Source	What it measures	How we computed it
GFF pseudogene feature type	Genes BV-BRC's own annotation pipeline explicitly labeled pseudogene in column 3 of the .PATRIC.gff	grep -c $'\tpseudogene\t' <genome>.PATRIC.gff
BV-BRC partial_cds	BV-BRC's own summary-stats field for incomplete/partial CDS calls	p3-all-genomes --eq genome_id,<id> --attr partial_cds,partial_cds_ratio
Panaroo invalid-gene count	CDS whose translated sequence contains an internal premature stop codon (checked by actually translating each CDS, not reading any annotation field)	Ran panaroo --remove-invalid-genes --refind-mode off, grepped its stderr for "invalid gene! ... Has stop: True", and counted per source GFF

Comparison table

Genome	Contigs	GFF pseudogene count	BV-BRC partial_cds	BV-BRC partial_cds_ratio	BV-BRC hypothetical_cds	Panaroo invalid-gene count (ours)
1282.2719	4	0	0	0	553	25
1282.3129	103	0	0	0	525	0
904315.3	144	18	0	0	566	0
904320.3	87	59	0	0	600	0
904330.3	56	47	0	0	439	0
(remaining 12 S. epidermidis genomes)	36–90	18–34	—	—	—	0
(all 73 S. argenteus genomes)	10–30	—	—	—	—	0 total (all 73 genomes)

How we found it

We were debugging a severe Panaroo pangenome-construction slowdown (see context below) and, while testing panaroo --remove-invalid-genes in isolation, found that Panaroo's own CDS-translation check flags 25 genes in 1282.2719 as invalid due to internal stop codons. Cross-checking against BV-BRC's reported fields for the same genome (p3-all-genomes --eq genome_id,1282.2719 --attr partial_cds,partial_cds_ratio) returned 0 for both, and the GFF3 file itself contains zero features labeled pseudogene.

Reproduction

# Genome: 1282.2719 (Staphylococcus epidermidis strain CSF41498)
# GFF: fetched via BV-BRC CLI, p3-genome-fasta / genome GFF export

# BV-BRC-reported fields (all 0, despite the genes below being demonstrably broken):
p3-all-genomes --eq genome_id,1282.2719 --attr genome_id,cds,partial_cds,partial_cds_ratio,hypothetical_cds
# → 1282.2719  2490  0  0  553

# Genes independently confirmed to contain an internal premature stop codon
# (verified by translating the CDS sequence directly, via Panaroo v1.5.1's
# `--remove-invalid-genes` check):
fig|1282.2719.peg.62, peg.134, peg.219, peg.273, peg.438, peg.627, peg.656,
peg.657, peg.672, peg.993, peg.997, peg.1410, peg.1434, peg.1714, peg.1722,
peg.1796, peg.1808, peg.1917, peg.2000, peg.2115, peg.2305, peg.2378,
peg.2465, peg.2478, peg.2481

Additional annotation-pipeline inconsistencies on this genome

• Missing ##sequence-region header lines in the GFF3 (present on every other genome we checked in this species).
• Contig identifiers behave like real NCBI accessions (CP030246 etc.) rather than the synthetic placeholder pattern (<genome_id>.con.NNNN) seen on other older-pipeline genomes — so this isn't simply "unassembled," it's specifically the pseudogene/partial-CDS annotation that's missing.

Why this matters

Consumers relying on partial_cds/pseudogene fields to filter low-quality or degenerate gene calls (e.g., for pangenome analysis, comparative genomics, or QC pipelines) will silently retain genomes with substantial numbers of non-functional gene calls that these fields were designed to flag. In our case, this caused a severe (>100x) performance degradation in a downstream tool (Panaroo) that trusts gene-level input and only detects the problem via a full CDS translation check — something most consumers of BV-BRC data won't run themselves.

Ask

Could BV-BRC confirm whether partial_cds/pseudogene calling was skipped or failed for this genome's annotation run, and whether other genomes from the same submission batch/pipeline version are similarly affected?

[jananiravi]

Update

Bug Report: BV-BRC metadata provides no signal for genomes carrying clustered translation-invalid CDS calls

Summary

Certain genomes (e.g., 1282.2719, S. epidermidis strain CSF41498) carry a disproportionate cluster of CDS features with internal premature stop codons — verified by direct translation, not by any BV-BRC-reported field. We could find no metadata field, consistency score, or quality flag that distinguishes these genomes from clean ones, including fields specifically designed to catch this class of problem.

What we checked and ruled out (across 3 species, ~390 genomes total)

Field	Result
owner	Same value (PATRIC@patricbrc.org) on both bad and good genomes in every species tested (68/73 clean S. argenteus genomes share it with the 2 bad S. epidermidis genomes)
assembly_method	Same tool/version (e.g. CLC Genomics Workbench v.6.2) on both bad and good Campylobacter genomes
sequencing_platform / sequencing_depth	Overlapping ranges, no separation
checkm_completeness / checkm_contamination	Bad genomes score 97-100% complete, 0-3% contamination — indistinguishable from clean genomes
coarse_consistency / fine_consistency	Same range (96-100%) for both groups
genome_quality_flags	Empty for every genome checked, bad or good
GFF pseudogene feature count	0 for the worst offender (1282.2719), despite it carrying 25 translation-verified broken CDS
partial_cds / partial_cds_ratio	0 for 1282.2719

What we found instead

Only a direct CDS-translation check (stop-codon scan) reveals the problem. Critically, it's not about total count — it's about concentration in a single genome:

• A 297-genome Campylobacter jejuni batch with 51 such genes spread across 20 genomes (max 7 in any one) processed normally in ~10 minutes.
• A 17-genome S. epidermidis batch with only 25 such genes, but all in one genome (1282.2719), degraded catastrophically (>100x slower, sometimes not completing) in downstream pangenome tools that trust per-genome gene calls.

Reproduction

Genome: 1282.2719 (Staphylococcus epidermidis strain CSF41498)

BV-BRC-reported fields (all 0, despite the genes below being demonstrably broken):

p3-all-genomes --eq genome_id,1282.2719 --attr genome_id,cds,partial_cds,partial_cds_ratio,hypothetical_cds
-> 1282.2719  2490  0  0  553

Genes independently confirmed to contain an internal premature stop codon

(verified by translating the CDS sequence directly, via Panaroo v1.5.1's

--remove-invalid-genes check):

fig|1282.2719.peg.62, peg.134, peg.219, peg.273, peg.438, peg.627, peg.656,
peg.657, peg.672, peg.993, peg.997, peg.1410, peg.1434, peg.1714, peg.1722,
peg.1796, peg.1808, peg.1917, peg.2000, peg.2115, peg.2305, peg.2378,
peg.2465, peg.2478, peg.2481

Ask

This isn't a request to fix one genome — it's that no existing BV-BRC quality signal (CheckM stats, consistency scores, quality flags) can be used by downstream consumers to detect or filter this failure mode. Would BV-BRC consider exposing a per-genome "CDS translation validity" metric (count/ratio of CDS with internal stops), independent of the pseudogene GFF annotation, so pipelines can screen for this before it silently breaks tools that assume clean gene calls?