FlaGs3

Conserved gene neighbourhoods around your proteins of interest.

Example output

The demo list holds 15 proteins annotated as the export chaperone SecB. FlaGs3 resolved 14 of them to a genome and drew these figures; every one is also written as PDF, and the tables behind them come in the archive.

FlaGs3 Neighbours figure of the demo list

One row per query. The query has a black outline and points right; four genes on each side are drawn to scale. Genes of one colour, carrying one number, are homologues of each other; grey genes have no homologue in any other row. Here the query (Q1) always sits behind family 1 and one of families 2, 3 or 13, while the genes further out change from genome to genome. The panel under the rows is the optional protein clusters legend: every family with its product names.

FlaGs3 Tree figure of the demo list

The same neighbourhoods in the classic look, in the order of a phylogenetic tree built from the query proteins, with branch support in red. Related queries with the same gene context end up next to each other.

FlaGs3 Domains figure of the demo list

Pfam domains drawn as numbered wedges inside the genes, named in the legend. Family labels take a G prefix here so they cannot be mistaken for domain numbers. A grey gene with a familiar domain is often the most useful thing on the page.

FlaGs3 Defence systems figure of the demo list

DefenseFinder and PadLoc run on the flanking genes; a system either tool calls is a band under the genes it covers. Here the toxin–antitoxin–chaperone system CmdTAC (D1) is called in 4 of the 14 neighbourhoods, on the genes next to the query.

FlaGs3 Mobile elements figure of the demo list

geNomad scans the sequence around each query for proviruses and plasmid signal. Nine of the 14 neighbourhoods lie inside a predicted provirus: Caudoviricetes (M1) in eight, Inoviridae (M2) in one.

FlaGs3 All-in-one figure of the demo list

Everything at once, on the tree: families, domains, and one lane of bands per tool. Every system has its own colour, the same in every figure of the run.

FlaGs3 Classic figure of the demo list

The original FlaGs figure: compact arrows with the family number inside.


Give FlaGs3 a list of protein accessions. For each one it fetches the genome, cuts out the flanking genes, clusters all flanking proteins into homologous families with jackhmmer, and draws the neighbourhoods side by side with families in matching colours, so that conserved gene context stands out. Optionally it builds a phylogenetic tree of the query proteins, annotates Pfam domains on every flanking protein, and looks for defence systems, proviruses, plasmid replicons and secretion systems around the queries. Results stay on the server for 5 days and can be downloaded as one archive.

This is the FlaGs3 web service; the command-line tool with all its options is on GitHub. The previous version runs as webFlaGs.


Query proteins

One query per line: a protein accession, or a protein accession and its assembly separated by whitespace. Up to 200 queries per job. Load the demo list.

The file replaces the text box when both are given.
Start from BlastP

Optional. Give up to 5 starting proteins and FlaGs3 finds their homologues itself: it searches RefSeq proteins without eukaryotes with BlastP and uses the best hits as queries, 50 in all, shared equally between the starting proteins. The hits join the accession list above, which may then be left empty. One address or email can have 3 BlastP jobs queued or running at a time.

An accession on a line of its own, or a protein sequence in FASTA or as bare residues; separate sequences with a FASTA header or an empty line. The hits and their E-values are in blast/hits.tsv of the archive.
Neighbourhood and clustering
MMseqs2 clusters the flanking proteins several times faster than jackhmmer and is the kind choice for long lists; jackhmmer finds more distant homologues. Left empty, the E-value is the method's own default (1e-10 for jackhmmer, 1e-3 for MMseqs2). Iterations apply to jackhmmer only.
Extras
mafft, trimAl and VeryFastTree; adds minutes for large lists. The nucleotide similarity graph compares the DNA of each neighbourhood with the next one in the tree (blastn) and draws the similar blocks as pale blue ribbons in the all-in-one figure.
tRNA, rRNA and other non-coding RNA get families like the proteins (nhmmer).
Every family's product names under the neighbors and all-in-one figures.
Adds digest.md to the report: the whole run as one compact text document.
Systems and mobile elements

Each tool adds its calls to the figures and a table to the report, and time to the job.

Anti-phage defence systems among the flanking genes.
Anti-phage defence systems; one both tools agree on is drawn once.
Proviruses and plasmid signal in the sequence around each query.
Plasmid replicon types on the contig carrying each query.
Secretion systems in the sequence around each query.
For geNomad and Sismis. PlasmidFinder always reads the whole contig; the defence tools read the flanking genes.
Needs geNomad switched on. Besides proviruses, which have coordinates and are always drawn, geNomad classifies the scanned sequence as a whole; this prints that call (plasmid, virus) beside the row. It can be a false positive: geNomad sees only the stretch set by the scan range, not the whole contig, so the call describes that stretch and can change with the range.
Run
Passed to NCBI with every request, as their Entrez service requires.
Names the result archive.

Input format

Each line is one query. A bare accession is looked up at NCBI, which finds the genome it comes from; a pair PROTEIN<tab>ASSEMBLY names the genome directly and skips the lookup. Lines starting with # are ignored. Three kinds of genome are understood:

  • NCBI: RefSeq or GenBank proteins (WP_…, NP_…, YP_…, AAC…) and assemblies (GCF_…, GCA_…). A bare accession is enough.
  • ENA: the protein as it is named in the ENA record, paired with a sequence accession (U00096, CP003200.1) or a WGS set (CABHOY01). Always a pair.
  • MGnify: a MAG locus tag paired with its catalogue accession (MGYG000454827_00001<tab>MGYG000454827). Always a pair.

Queries that cannot be resolved or whose genome cannot be fetched are listed in the run summary and the job still completes for the rest.

Contact

Questions, problems and ideas are welcome at fedor.taratorkin@med.lu.se.