interproscan: staged database is ignored (relative INTERPROSCAN_CONF, wrong bin/data directories) #13009

Description of the bug

interproscan ignores the database passed as interproscan_database and always uses the small sample database built into the container. With a real database, the run fails:

Running hmmpress (/usr/local/bin//hmmpress) on data/pirsf/3.10/sf_hmm_all
failed to open SSI index data/pirsf/3.10/sf_hmm_all.h3i
Could not run hmmpress!

Cause

This is checked against quay.io/biocontainers/interproscan:5.59_91.0--hec16e2b_1.

1. interproscan.sh runs cd "$(dirname "$INSTALL_DIR")", which moves into /usr/local/share/InterProScan, before it checks INTERPROSCAN_CONF.
2. The module sets export INTERPROSCAN_CONF=interproscan.properties as a relative path. After the cd, that resolves to the container's own /usr/local/share/InterProScan/interproscan.properties, not the edited copy in the task work dir.
3. The container's properties file has data.directory=data and bin.directory=bin, both relative to the install dir. InterProScan therefore reads the bundled sample data. That data is not pressed and sits on a read-only image, so hmmpress fails. The staged data/ folder is never used.
4. Separately, the sed sets bin.directory=$INTERPROSCAN_DIR/bin, which is /usr/local/bin. The member-database binaries live in /usr/local/share/InterProScan/bin, for example bin/pirsf/3.10/pirsf.pl. data.directory is never rewritten either.

The tests pass because the test profile only runs applications that work with the bundled sample data.

Suggested fix

if [ -d 'data' ]; then
    INTERPROSCAN_DIR="$( dirname "$( dirname "$( which interproscan.sh )" )" )"
    INTERPROSCAN_PROPERTIES="$( find "$INTERPROSCAN_DIR/share" -name "interproscan.properties" )"
    cp "$INTERPROSCAN_PROPERTIES" .
    sed -i \
        -e "s|^bin\\.directory=.*|bin.directory=$( dirname "$INTERPROSCAN_PROPERTIES" )/bin|" \
        -e "s|^data\\.directory=.*|data.directory=$PWD/data|" \
        interproscan.properties
    export INTERPROSCAN_CONF=$PWD/interproscan.properties
fi

Also worth documenting: interproscan_database must be the data/ folder of a matching InterProScan release, meaning 5.59-91.0 for the current container, and its HMMs must already be pressed with python3 setup.py -f interproscan.properties. Otherwise hmmpress runs inside every task and parallel tasks race.

Workaround

Bind the database over the container's data folder:

process {
    withName: '.*:INTERPROSCAN' {
        containerOptions = '-B /abs/path/interproscan-5.59-91.0/data:/usr/local/share/InterProScan/data'
    }
}

System information

• Module: modules/nf-core/interproscan (current master)
• Container: interproscan:5.59_91.0--hec16e2b_1
• Executor: Slurm, Singularity
• Found via INTERPROSCAN fails with real database: version mismatch and nested data directory proteinannotator#114