Plasmids¶

Stage 06.plasmids asks which contigs of the finished genome are plasmids rather than chromosome. It runs in all four modes, and the caller reads only contigs_final.fasta, so the call itself does not depend on which front end produced the assembly. The supplementary BLAST line beside each call does — see below.

Platon v1.8 is the caller and always runs. geNomad v1.12.0 is an optional second opinion: it is switched on from the phage stage, not from here, because one geNomad run calls viruses and plasmids together.

flowchart LR
    F([contigs_final.fasta])
    P[plasmid_search<br/>Platon]
    VP([06.plasmids/sample/platon/<br/>verified_plasmids.txt])
    F --> P --> VP

    GN([07.phages/genomad/sample/<br/>..._plasmid_summary.tsv<br/>only when geNomad is opted in])

    PC[plasmid_concordance]
    VP --> PC
    GN --> PC
    PC --> OUT([sample_plasmid_concordance.tsv])

Which of the two end files is the deliverable depends on one config key:

`phage.caller`	Plasmid deliverable	geNomad
`virsorter2` (default)	`06.plasmids/{sample}/platon/verified_plasmids.txt`	never runs
`genomad` (opt-in)	`06.plasmids/{sample}/{sample}_plasmid_concordance.tsv`	runs once per sample, in stage 07

Platon runs either way. See Prophages for the caller choice itself.

Terms used on this page


FDR	false discovery rate, the expected share of calls at a given score that are wrong

Platon¶

Rule plasmid_search.

Platon scores every contig on its protein content. Each protein family carries a replicon-distribution score (RDS) — how much more often that family is seen on a plasmid than on a chromosome — and the contig's summed score decides the call. It needs the Platon database, set as directories.platon_db; see Reference databases.


In	`contigs_final.fasta`, plus a BLAST table computed over those same contigs (which one depends on the mode — see below)
Out	`06.plasmids/{sample}/platon/` — `contigs_final.tsv`, `contigs_final.plasmid.fasta`, `contigs_final.chromosome.fasta`, `verified_plasmids.txt`
Next	the concordance rule (when geNomad is on), and the mobilome module's replicon table (when it is on)

Platon's own output files are named after the input, which is always contigs_final.fasta, so the stem is contigs_final in every mode.

The BLAST line beside each call¶

For every contig Platon calls a plasmid, the rule takes that contig's first line in the BLAST table and asks — case-insensitively — whether it mentions "plasmid". In practice that is the subject title, which is why Decontamination warns against reordering the -outfmt columns. The answer is one line per contig in verified_plasmids.txt:

sample01: contig_3 is a plasmid.
sample01: contig_9 was not verified by BLAST search.

A genome Platon called no plasmid on gets a single Platon found no plasmid in sample … line instead. That is a result, not an error.

The BLAST line is an annotation, never a filter — nothing is dropped on the strength of it. Both signals can be fooled by the same thing: a mobile element sitting on a genuinely chromosomal contig pushes Platon's score and that contig's best nt hit in the same wrong direction, so agreement here is weaker evidence than it looks.

Which BLAST table is used differs by mode, and is resolved once at startup:

Mode	Table	Why
`illumina`, `contigs`	the decontamination screen's own BLAST output	the screen already ran on this exact contig set
`nanopore`, `hybrid`	a second megablast over `contigs_final.fasta` (rule `blast_final_contigs`)	nothing guarantees Medaka preserves contig headers, and the hybrid screen runs on the Illumina draft, whose SPAdes `NODE_…` names can never match Flye's `contig_…` names

The lookup is by contig ID, so the wrong table would report every plasmid as "not verified by BLAST search" — silently, with no error. That is the reason it is resolved centrally rather than per rule.

An almost-empty Platon directory is a normal result, not a failure

Platon refuses to classify a contig longer than 500 kb (MAX_CONTIG_LENGTH in Platon's own source). A closed chromosome is therefore in neither the plasmid nor the chromosome FASTA — it was never assessed. On a complete Nanopore or hybrid assembly this is the usual case, and the per-contig table can legitimately be header-only.

A real Platon crash looks different: the rule traps the exit status and writes one line, {sample}: Platon exited with status N; see the log., into verified_plasmids.txt instead of killing the sample. That line is the only reliable signal that Platon failed, and both downstream readers key on it.

geNomad as a second opinion¶

Rule plasmid_concordance, defined only when phage.caller: genomad.

geNomad classifies contigs from gene content against its own marker set — a different method with different failure modes — so genuine agreement between it and Platon is much stronger evidence than Platon agreeing with a BLAST screen that was run for another purpose. geNomad is licensed by Berkeley Lab for academic / non-commercial use; the terms of every tool are listed under Licensing.

The rule joins four files into one table: Platon's per-contig table, Platon's chromosome FASTA, verified_plasmids.txt, and geNomad's plasmid summary from stage 07. One row per plasmid candidate — the union of the two tools' plasmid calls. Contigs both tools agree are chromosome are left out.

Column	Holds
`sample`, `contig`	the join keys
`platon_call`	`plasmid` \| `chromosome` \| `not_called` \| `not_assessed`
`platon_rds`	Platon's replicon-distribution score, or `NA`
`platon_blast_hit`	`hit` \| `no_hit` \| `NA` — the supplementary BLAST line above
`genomad_call`	`plasmid` \| `absent`
`genomad_score`, `genomad_fdr`	geNomad's plasmid score, and its FDR when calibration ran (usually `NA`)
`agreement`	which tools called it
`confidence`	`high` \| `medium` \| `low`, from `agreement` alone

Agreement drives the tier, and nothing is discarded — a disagreement is flagged and kept, so the table is its own audit trail:

Platon	geNomad	`agreement`	`confidence`
plasmid	plasmid	`both`	high
plasmid	—	`platon_only`	medium
not called	plasmid	`genomad_only`	medium
chromosome	plasmid	`conflict`	low
crashed	anything	`platon_unavailable`	low

Three rows from the Klebsiella pneumoniae ATCC BAA-2146 positive control (sample and genomad_fdr not shown). The two tools agree on all three plasmid contigs that reached this stage — the genome's fourth plasmid never did, because decontamination had already removed it (see below):

contig          platon_call  platon_rds  platon_blast_hit  genomad_call  genomad_score  agreement  confidence
NZ_CP006661.1   plasmid      31.8        hit               plasmid       0.9925         both       high
NZ_CP006662.2   plasmid      20.1        hit               plasmid       0.9937         both       high
NZ_CP006663.1   plasmid      27.1        hit               plasmid       0.9943         both       high

The comparison is not symmetric yet

The rule reads only geNomad's plasmid summary, which lists positive calls. So genomad_call: absent means "geNomad did not call this a plasmid" and cannot distinguish "it actively called it chromosome" from "it never scored it". The forward clash — Platon chromosome against geNomad plasmid — is flagged conflict / low. The reverse case, Platon plasmid against geNomad silence, is reported platon_only / medium rather than as a conflict. Nothing is hidden either way; only that one tier label is conservative.

If both tools called plasmids but none of their contig IDs match, the script says so in the log. The join has broken — a header not trimmed to its first token somewhere upstream — and every genuine both / high has split into two single-tool medium rows.

What the mobility ladder takes from here¶

When the mobilome module is on, it reads this stage rather than re-detecting anything. Platon already answers both questions the top of the ladder needs: whether a contig is a plasmid at all, and whether that plasmid carries the machinery to move itself, from its own # Conjugation, # Mobilization and # OriT counts.

Platon counts	Plasmid mobility	Ladder
conjugation genes present	`conjugative`	tier 6 — predicted self-transmissible
no conjugation, but a relaxase or an oriT	`mobilisable`	tier 5 — needs a helper
neither	`non-mobilisable`	—

Conjugation is tested first because a relaxase is part of a conjugative system, not an alternative to one. Contigs Platon skipped for length are called chromosome only at 2 Mb and above; between 500 kb and 2 Mb the replicon is reported unknown, because megaplasmids of that size exist. When geNomad is on, the concordance table is passed in as well, so a plasmid Platon missed does not silently become "chromosomal, intrinsic candidate".

Before you conclude a plasmid is absent¶

Decontamination can remove a small plasmid before this stage sees it

The auto selector keeps contigs whose assigned genus matches the sample's dominant genus. A broad-host-range plasmid can be discarded precisely because it is mobile and its closest database relatives sit in another genus — the measured case is the 2,014 bp plasmid pMYS of the control genome above, dropped as Escherichia (Decontamination). Check 02.assembly/{sample}/contaminants/contig_taxonomy_decisions.tsv before trusting an absence. In hybrid mode there is a second route, worse than the first: a contig dropped there takes its long reads with it, so the sequence is missing from the assembly and not merely from the plasmid call (Route 2).

Small plasmids are also lost upstream of that, at read filtering and assembly. The settings that decide it are covered on Nanopore mode.

What to check afterwards¶

06.plasmids/{sample}/platon/verified_plasmids.txt — the per-contig call and its BLAST line, or the one-line Platon failure message.
06.plasmids/{sample}/{sample}_plasmid_concordance.tsv — when geNomad is on, the tiered table, including every flagged conflict.
logs/plasmid_search_{sample}.log — Platon's own verbose output.
logs/plasmid_concordance_{sample}.log — the join summary: candidate contigs, how many scored both, how many conflicts, and how many geNomad IDs matched a Platon contig ID.