Help

Input

Protein or nucleotide sequences in FASTA or FASTQ format, pasted or uploaded (plain, gzip, bzip2 or xz). Nucleotide sequences (reads, contigs) are translated in six frames: an open reading frame that follows a stop codon starts at its first methionine; one that runs off the start of the sequence is kept whole, as a possible partial precursor. Identical sequences are reported once, with their number of copies.

Reading the results
Prediction and P
The most probable class, a gene superfamily or NONE (not a conopeptide), and its probability. Probabilities are calibrated on held-out data: among predictions with P around 0.9, about 9 in 10 are correct.
P(NONE)
The probability that the sequence is not a conopeptide.
Best family if conopeptide
When you already know the sequence is a conopeptide (for example a peptide identified in the venom), use this column: the most probable superfamily once NONE is excluded, with its probability renormalised over the superfamilies.
Regions
Signal peptide (SIG), propeptide (PRO), mature peptide (MAT) and C-terminal propeptide (POST), from an alignment to the predicted superfamily's precursor model, with cleavage sites refined by position weight matrices. A mature peptide ending the precursor with a trailing glycine (optionally followed by basic residues) is predicted to be C-terminally amidated, and these residues are removed from the mature peptide.
Framework
Cysteine pattern of the predicted mature peptide (e.g. CC-C-C).
Limits
  • Short and divergent mature peptides are the hardest case; check their probabilities.
  • Superfamilies absent from the database cannot be predicted: such sequences get a low maximum probability or NONE.
  • The C-terminal propeptide (POST) is rarely detected.
  • By default, sequences of 30 residues or more are scored only if they pass HMMER's acceleration filters for some profile (much faster on transcriptomes; on the evaluation set no conopeptide was lost). Use "Score every sequence" to disable this.
Command line

For large or repeated analyses, install ConoDictor locally: pip install conodictor (needs pftools; see GitHub). The web server runs exactly the same program.