Skip to content
Documentation hub
READMEContributingAgent guideSecurity policyChangelogNotice & creditsnf-claw architectureChaining pipelinesEngine and version compatibilityPrepare DRAMP for Funcscan 4.0.0Known issues and troubleshootingPublication policy for the maintenance workflowsnf-claw over MCPValidation scope and samplesheet generationReplaying and verifying runsScientific output contractsRecorded workflow-source correctionsTimeouts, long analyses and monitoringUpdating and discoveryEnd-to-end validation

Generated from

docs/dramp-reference.md
3 min read View source
On this page6

Prepare DRAMP for Funcscan 4.0.0

The pinned Funcscan release uses AMPcombi 3.0.0. Its DRAMP downloader passes pandas NaN values from missing sequence cells to a string regex and crashes. APD is unaffected. An explicit prepared reference avoids this downloader and records the database actually used. The pinned pipeline and container code remain unchanged.

Navigation

Overview • Download • Prepare the reference • Build the index • Run Funcscan

Overview

Download the complete official general-AMP TSV and retain it beside your records. The endpoint is mutable: its URL is not a version or a checksum.

Downloading the general AMP dataset

For example, with a recent curl:

bash
curl --fail --location --http1.1 --remove-on-error --connect-timeout 15 --max-time 300 \
  --output general_amps.txt \
  'https://dramp.cpu-bioinfor.org/downloads/download.php?filename=download_data/DRAMP3.0_new/general_amps.txt'

Check the transfer succeeded and the response length/checksum against your intended snapshot. Do not prepare a partial download. The server validation snapshot on 2026-10-27 was 13,221,293 bytes, SHA-256 4e044b3a4d9fcfc24cff81cbcf9fdb48184b41d8773837c697a2f205c1ab9dd5. That snapshot contains no missing sequences, so its downloader succeeds today; an explicit blank-cell control reproduces the latent failure in the actual released container.

Preparing the reference directory

Use a fresh destination under an existing parent:

bash
nfclaw prepare-dramp /data/general_amps.txt --outdir /refs/amp_DRAMP_database

This standalone command needs no pipeline checkout or optional Python packages. It writes general_amps.fasta, general_amps.txt and preparation.json, and prints the JSON record. The directory is private (0700), its files are private (0600), and existing destinations are refused. Its name must contain the case-sensitive substring DRAMP: AMPcombi selects its annotation parser from that name.

The explicit identified-uppercase-canonical-20-v1 policy retains a row only when it has a DRAMP identifier and a nonempty sequence made entirely of uppercase ACDEFGHIKLMNPQRSTVWY. It preserves the exact retained sequence, identifier, every annotation value, column and row order. It never trims, uppercases, renames or imputes data. Missing identifiers/sequences and other sequences are excluded with reason counts and bounded example IDs. This is compatibility with the consumer's canonical-sequence criterion; other peptides can be biologically valid. Unsafe or duplicate nonempty IDs, malformed/duplicate/missing columns and no retained sequences cause refusal. The held raw input is hashed before and after parsing; changed input is refused before publication. These snapshots do not prove continuous immutability.

The validation snapshot contains 12,784 rows: 10,564 retained, 2,219 noncanonical sequences and 1 missing identifier excluded. All retained annotation values and FASTA identifiers/sequences were independently checked against the complete raw file. The raw file is not overwritten. Review the counts for your own snapshot rather than assuming these numbers stay current.

Building the MMseqs2 index

Build the reference index before analysis. The preparation record describes the earlier FASTA/annotation stage and explicitly says its index is not built. Use the tested image digest:

bash
docker run --rm --network none --user "$(id -u):$(id -g)" \
  --mount type=bind,src=/refs/amp_DRAMP_database,dst=/ref \
  quay.io/biocontainers/ampcombi@sha256:53a73be7a84bb1e921c2bbbd81d6d48453b0b3ab5d0ae1e6709c0d6dd662c237 \
  sh -eu -c 'mkdir /ref/.mmseqs2-building
    mmseqs createdb /ref/general_amps.fasta /ref/.mmseqs2-building/ref_DB -v 0
    mv /ref/.mmseqs2-building /ref/mmseqs2'

Run this once on a freshly prepared reference, with no concurrent writer. If it fails, do not supply the incomplete directory to analysis: AMPcombi treats the existence of mmseqs2 as proof that indexing is done. Require the command to succeed and the eight nonempty index files: ref_DB, ref_DB.dbtype, ref_DB.index, ref_DB.lookup, ref_DB.source, ref_DB_h, ref_DB_h.dbtype, ref_DB_h.index under mmseqs2/.

Execution with Funcscan

Supply your normal input and screening options, selecting DRAMP explicitly:

bash
nfclaw run funcscan --input /data/samplesheet.csv --outdir /results/funcscan \
  --nxf-ver 25.10.4 --amp-ampcombi-db-id DRAMP \
  --amp-ampcombi-db /refs/amp_DRAMP_database --run-amp-screening true

The standard bundled demo selects APD. To test DRAMP with demo input, pass these same database options with --demo. AMP predictors, ARG/BGC/CAZyme/taxonomy branches and their databases remain separate choices. A completed selected AMP branch does not validate every Funcscan branch.

The pinned container built the complete index and recovered an 84-residue known-answer query as the exact DRAMP00222 sequence with its annotation. That consumer check left the full prebuilt reference unchanged. It establishes reference ingestion/search/annotation behavior, not biological AMP prediction accuracy. Normal run provenance checksums the supplied directory including indexes; keep those files available and unchanged for guarded replay.