Quality checks and flags

After RiboPilot annotates a contig (see Get started), it runs a set of automatic quality checks and raises a curation warning for anything that looks unusual. Warnings never change your assemblies or annotations. They flag something that may need your attention.

You can find the results in three places:

• the Checks column of the Assemble and Annotation Review tables, which summarizes every check
• the Warnings column of Annotation Review, which counts the warnings on the contig as a whole
• the Curation warnings panel in the Annotation Review contig details window, lists the contig’s warnings and each feature that has a warning

Below is a list of all the checks and warnings columns you will see in the app:

Column	Where	What it holds	Explained in
Checks	Assemble table (per sample); Annotation Review table (per contig)	Every check (per sample or per contig) on one line, worst first, colored by the worst finding.	The Checks column
Warnings	Annotation Review table	How many curation warnings the contig has.	Curation warnings
Screen	Assemble details window (contig table); Annotation Review table	The BLAST screen verdict: on target, off target, or too little evidence to say. The Screen why column next to it on Assemble gives the reason.	The BLAST screen
Chimera	Assemble details window (contig table); Annotation Review table	The chimera check level: severe, minor, or clean.	Chimeric contigs
ITSx Warnings	Assemble details window (contig table); Annotation Review table	What the standard ITSx check finds for the contig, for example itsx_chimeric, itsx_no_5.8S, or itsx_boundary_shift(...). Gray “not run” means ITSx gave no answer; gray “not checked” means no ITSx check is stored yet. Checks lists both as “ITSx not checked”.	GenBank submission
Feature BLAST warnings	Annotation Review table	Warnings about weak or off-target BLAST hits for each feature.	Feature BLAST

This page explains the Checks column first, then each family of checks in turn, and ends with the per-group length and identity tables.

The Checks column

The Checks column puts every check on one line, worst first, in a color that matches the worst finding. On the Assemble table it summarizes all of the sample’s kept contigs.

Checks says	Color	What happens
off target, contaminant, severe chimera, or ITSx chimeric	red	Lock and the Export window ask first, until the contig is marked Reviewed.
minor chimera, uncertain screen, ITSx, ITSx not checked, warnings, or feature flags	amber	Worth a look, but nothing asks when you Lock or Export.
ok	green	The best BLAST matches are in your sample’s phylum, and nothing is flagged.
no reference	gray	Nothing is flagged, but the BLAST screen had no usable match to judge by.

A red finding does not stop you, but locking or exporting that sample will ask you to confirm. Setting a problematic sample as Mark Reviewed will override those prompts on lock or export (see Annotation Review).

Curation warnings

In the Annotation Review module, these are the warning codes counted by the Warnings column and reported in the Curation warnings panel when you examine a contig in detail.

Warning	Level	What it means	What to do
missing_18S, missing_5.8S, missing_28S	contig	One of the three rRNA genes was not found.	Expected when the contig contains only part of the nuclear rDNA complex. Otherwise look at Coverage and windowed BLAST for a break.
missing_ITS1, missing_ITS2	contig	A spacer was not found between its two genes.	Usually follows a missing gene. If both genes are there, check the boundaries in Edit active feature boundaries.
out_of_order	contig	The features are not in the order 18S, ITS1, 5.8S, ITS2, 28S.	Possible assembly error. Export marks it misassembly. Check the feature map and read coverage before you submit it.
overlap	both	A feature overlaps the one before it.	Move the boundary in Edit active feature boundaries. Export marks it misassembly.
gap	both	Unannotated sequence between two features, more than 5 bases long.	Check the boundaries, and extend the gene or spacer if the gap is part of it.
wrong_strand	both	A feature lies on the opposite strand from the rest.	Likely a misassembly or a wrong call. Check the feature model.
length	feature	The feature is shorter or longer than its group’s expected range (see per-group tables).	Common for the spacers, and often fine. Check that the boundaries look right.
truncated	feature	Less than 90 percent of the gene’s rRNA model was recovered.	Nothing: the export writes the gene as partial.
low_identity	feature	The gene falls below its group’s identity floor (see per-group tables).	An assembly error, or a group with few other rRNA references. Look at the Reference alignment.
contaminant	contig	The contig’s best BLAST hit is from a different phylum than your sample.	Red in Checks. Check the contig before you lock it: it may be another organism’s rDNA.
contaminant_unchecked	contig	The contamination check could not run: the BLAST screen has no usable hit, or your sample’s lineage is unknown.	If NCBI could not be reached, run the sample again once it can (see Troubleshooting). If the sample has no taxon, set one in the Taxonomy step.
not_nuclear_rdna	contig	The best BLAST hit is not a nuclear rDNA record, for example a mitochondrial or bacterial rRNA.	The contig may not be nuclear rDNA. Check its features and hits before you keep it.
no_clade_ruleset	contig	Your sample’s group has no ruleset of its own, so the generic animal ranges were used.	Nothing. See How the checks are tuned per group.

The BLAST screen and auto-triage

Every contig is run through a BLAST search once it is assembled. The result feeds the Screen column, the contamination warnings above, the chimera check, and the triage rules below, which propose whether to keep or drop the contig.

What the screen searches

By default each contig is searched against the bundled rDNA database inside the container: about 600,000 animal nuclear rDNA records from GenBank. This database also includes NCBI’s 16S, eukaryote small subunit, and eukaryote large subunit rRNA sets so that a contaminant bacterial, fungal, or plant contig still finds a match. Contigs that match nothing in the bundled database are searched against the NCBI nucleotide database (core_nt) over the internet.

To change how the BLAST search works, open a sample’s BLAST Options in the Assemble module and pick a Search choice: “Bundled database only” never goes to NCBI, and “NCBI only (core_nt)” skips the bundled database.

In Annotation Review, the Hit Source column says which database was used to determine the contig’s best BLAST hit: “bundled”, “NCBI”, or “seed file” (the assembly’s own seed database, which is only a hint).

How contigs are kept or dropped

Once a contig has been screened, RiboPilot checks the eight rules below in order and proposes to keep or drop the contig. For rules whose action is Flag, the contig status stays undecided and the Why column in the sample’s contig table gives the reason. You can override any call by hand with the K (keep) and D (drop) buttons in the Assemble details window.

#	Rule	What it means	Default action
1	Annotation has not run	Neither annotation tool has reported on this contig yet.	Flag only
2	BLAST screen has not run	No usable BLAST hit is recorded, or the only matches are whole-genome assemblies, so the contig cannot be judged. Also raises contaminant_unchecked.	Flag only
3	No annotations, non-rDNA hit	Annotation found no rRNA or ITS feature, and the best match is not a nuclear rDNA record.	Drop
4	No annotations but rDNA hit	Annotation found no rRNA or ITS feature, but the best match is a nuclear rDNA record.	Flag (can be set to Drop)
5	Has annotations but non-rDNA hit	The contig has an rRNA or ITS feature, but the best match is not a nuclear rDNA record.	Flag (can be set to Drop)
6	GenBank taxonomy disagrees with the sample	The contig’s BLAST hits agree on a taxon outside your sample’s phylum. This is the contaminant warning.	Flag (can be set to Drop)
7	Disagrees with sibling contigs	This contig’s top hit disagrees with the majority of this sample’s other contigs. A hint, not proof.	Flag only
8	Chimeric assembly (severe)	Parts of this contig match organisms from different phyla or kingdoms.	Flag only

Rule 6 compares at phylum rank. It catches a fungal, bacterial, or plant contaminant, but cannot separate two organisms in the same phylum.

To change what a rule does to a contig, select samples in the Assemble table and click the Triage button at the top. The Triage window lets you set each rule to Off, Flag, or Drop, shows how many contigs each rule would decide, and reruns triage for the selected samples when you click Apply.

Chimeric contigs

A chimera is one assembled contig containing sequences that do not belong together, such as rDNA from two organisms or two different copies of the repeat from one genome. Each contig gets a chimera level, shown in the Chimera column of Annotation Review and in the Checks column:

Level	What it means	Triage	Export
severe	Two parts of the contig match different phyla or kingdoms, each at 90 percent identity or more.	Rule 8 flags the contig.	Red in Checks. The submission_risk column says chimeric(...).
minor	Weaker evidence: windows match more than one class, genes agree only at a high rank, a ribotyper warning, or read evidence alone.	No rule fires.	Amber in Checks. Listed in the chimera_warning column of the summary CSV; not a submission risk.
clean	Nothing found.	No rule fires.	Nothing.

In the test project, CHIM01 is severe: the contig is an artificial chimera composed of sequence from ray-finned fish and Penicillium mold.

A real chimera in the sequence viewer: the origin track switches lineage at the junction

How the check works

RiboPilot looks for chimeras in four ways:

• Gene by gene. The 18S, 5.8S, and 28S genes each get their own best GenBank matches. Ideally these agree on one lineage.
• Along the contig. The contig is divided into 150 bp windows and each is searched against the bundled rDNA database. Neighboring windows that match the same taxonomic class are joined into segments, and neighboring segments that hit different classes suggest a chimeric join. Neighboring segments with different taxonomic assignment are only flagged when both segments span at least 450 bp (for class mismatches) or 300 bp (for phylum or kingdom mismatches).
• Gene structure. ribotyper, NCBI’s rRNA checker, tests that the 18S and 28S each look like one whole gene. ITSx’s own chimera call adds a little weight.
• Reads. Many mapped reads that stop at the same point, or a large shift in read depth, can indicate a chimeric join.

How to judge a flag

In the Coverage and windowed BLAST panel:

• BLAST origin shows segments with consistent taxonomic identity
• BLAST identity shows the best hit’s percent identity
• vertical red dashed lines mark possible chimeric junctions

The Windowed BLAST details panel below provides more details about the BLAST results.

A real chimera shows a clean switch from one lineage to another, often accompanied by a shift in read depth. RiboPilot already discounts the common false positives: windows in conserved stretches where many groups match equally well are shown as “unresolved” and changes in read depth at a gene boundary or at the ends of the contig are ignored. Below, the first contig of SRR33849683 is “minor” because its genes agree only at phylum level, although every window that can be placed matches a ray-finned fish:

Windowed BLAST details for a weaker signal that needs a look

If in doubt, copy the contig with Copy FASTA and run your own BLAST searches on the NCBI website.

Feature BLAST: each feature searched on its own

The contig BLAST search can look fine even if one feature is poorly assembled or contaminated. So RiboPilot also searches each core feature (18S, ITS1, 5.8S, ITS2, 28S) on its own against the bundled rDNA database.

The Feature BLAST warnings column in Annotation Review lists only the features that need a look:

Warning	Meaning
no hit	Nothing in the database matched the feature.
non-rDNA hit	The best match is not a nuclear rDNA record.
off-target (group)	The best match shares nothing with your sample at phylum level or below.
no BLAST, feature too small	The feature is under 50 bp, too short for a useful search. Features of 50 to 99 bp are searched but marked low confidence, because short pieces match many organisms equally well.
weak hit	Only the more sensitive dc-megablast retry found a match; megablast found nothing.

In the Annotation Review table, open a contig Details window to see the full Feature BLAST panel. This shows each feature’s best match, how closely it matches your sample (species, genus, family, and so on), its identity and coverage, and the algorithm that found it (megablast, dc-megablast in amber, or blastn for short features). Expand a row for the top 10 matches. “Not run” means the database was not available.

In the test project, CHIM01’s 28S comes from a Penicillium mold, so its best match shares nothing with the fish sample and the feature is off target:

Feature BLAST for CHIM01: the 18S matches the fish, the 28S is off target

Several contigs from one organism

When a sample assembles more than one rDNA contig, RiboPilot compares their 18S, 5.8S, and 28S genes. The result is shown in the Variant column of the Assemble step’s contig table. Too little shared gene sequence gives “not comparable”.

An isomer is RiboPilot’s label for contigs whose rRNA genes are nearly identical but which differ in the ITS sequence or flanking regions.

A paralog means the genes of the contigs differ. That can be real variation within one genome (known in octocorals and jellyfish), contamination, or a pseudogene.

A contaminant from the same phylum passes the contig screen, so check each variant’s Feature BLAST warnings before you lock one. Handle isomers and paralogs with care: consult the literature for your group, or examine a reference genome, to see whether rDNA variation is a known phenomenon.

How the checks are tuned per group

The expected lengths and identity floors depend on the group of organisms. RiboPilot ships five rulesets, with ranges taken from GenBank records:

Ruleset	Used for samples in
actinopterygii	ray-finned fishes
asteroidea	sea stars
scyphozoa	true jellyfish
octocorallia	octocorals
metazoa_generic	any other group; the contig gets the no_clade_ruleset note

To use another one, pick it as Clade ruleset in the sample’s Curate Options (Assemble step). Leaving metazoa_generic there lets RiboPilot choose from the lineage.

Expected lengths, in bases. A length warning means a feature fell outside its range, which is common and often fine for the spacers:

Feature	Generic animal	Ray-finned fishes	Sea stars, jellyfish, octocorals
18S	1600 to 2100	1750 to 1950	1700 to 1900
5.8S	140 to 180	140 to 175	150 to 170
28S	2000 to 6000	2500 to 5000	2000 to 6000
ITS1	120 to 1600	150 to 1200	120 to 1200
ITS2	120 to 1600	150 to 1200	120 to 1200

Identity floors, in percent. Below these a gene gets low_identity:

Gene	Generic animal	Ray-finned fishes	Sea stars, jellyfish, octocorals
18S	95	97	93
5.8S	93	95	90
28S	80	85	78

The floors are lower for sea stars, jellyfish, and octocorals because GenBank has fewer rDNA records for these groups. All of these are working values, not published standards: a length or low_identity warning is a prompt to look, not a verdict. In every ruleset, a gene is truncated when less than 90 percent of its rRNA model was recovered, identity is judged only when the reference covers at least half of the gene, and annotation gaps of 5 bases or fewer are closed automatically.

Want ranges for your own group? Open an issue describing it. Until then, the generic animal ruleset works for any group; it is just more lenient.