GET_FINAL_METABOLITES_STATIC took chembl_db as a `path` input, so Nextflow
copied the ~9.2 GB SQLite DB into each task's work directory. With ~29
concurrent tasks that is hundreds of GB of I/O, against a process that also
declared only 1 GB of memory. Every task failed with exit 1 and retried ten
times, producing 746 errors and an empty 1b_final_metabolites/ output.
Pass the DB as a `val` path instead so tasks read it in place from the
dreamdock-data PVC, and give the process parameterised growing memory.
Source-only snapshot of the cluster branch for WES execution. Large
reference files (HPA/MANE/ensemble FASTA, model weights, ~597 MB) are
omitted: they are baked into the container images at build time and
mounted from the dreamdock-data PVC at runtime, and exceed the Gitea
request size limit.
Pipeline entry point is main.nf, which orchestrates the biotransformer,
conplex and tissue modules as a single workflow. Ligand inputs are read
from the eureka workspace; protein_zarr and chembl_db come from the
dreamdock-data PVC.
