biocomposer 0.1.4

LLM-wired bioinformatics pipeline orchestration over containerised tools

biocomposer

Graph-based orchestration of bioinformatics tools. Full documentation: https://mlberkeley.github.io/BioComposer-graph-based-orchestration-of-bioinformatics-tools-documentation/

Setup (both local and Modal)

pip install biocomposer
mkdir inputs              # put your FASTA/PDB/etc. files here
mkdir run                 # holds your pipeline script

Create a .env file in your working directory with the API key(s) you use (GEMINI_API_KEY, GOOGLE_API_KEY, ANTHROPIC_API_KEY, or OPENAI_API_KEY).

Create run/run_pipeline.py to define your pipeline.

Running locally

Prerequisites: Docker installed and running, bv installed (cargo install biov).

biocomp run run/run_pipeline.py --env .env

Outputs go to ./results/.

Running on Modal

pip install modal
modal setup
biocomp run --modal run/run_pipeline.py --env .env
modal volume get biocomp results/fasttree_output_1/stdout/fasttree_stdout.txt ./fasttree_stdout.txt

Flags

--env <path>      [both]  env file with API keys (default .env)
--clean           [both]  clear previous outputs first
--modal           [both]  run on Modal instead of locally
--gpu T4          [modal] GPU type (default A10G)
--memory 32768    [modal] sandbox memory in MB (default 8192)
--shell           [modal] drop into an interactive sandbox shell

Overriding a tool's command

By default a node runs the tool's entrypoint binary with its registry args. Override either when needed (slots like {input}/{output} map to the node's I/O):

# run a different exposed binary instead of the entrypoint
g.add_node("trimal", entrypoint_override="readal")

# replace the argument template
g.add_node("colabfold", args_override="--num-recycle 0 {fasta} {output_dir}")

# both together (common for non-entrypoint binaries)
g.add_node("trimal", entrypoint_override="readal",
           args_override="-in {alignment} -out {trimmed} -fasta")