jn_cuclark

jn_cuclark is a GPU-accelerated metagenomic classification package for Jetson Nano-class systems. It combines the CUDA-enabled cuCLARK core with two C++ front ends:

• kent: single-node orchestration for database setup, classification, abundance estimation, and report generation
• kent-mpi: MPI-based coordination for multi-node runs

This repository contains the software stack. Reference databases and sequencing reads are not bundled here.

Components

• src/: CUDA/C++ implementation of cuCLARK, cuCLARK-l, and helper binaries
• app/kent.cpp: single-node CLI wrapper around the shell scripts in scripts/
• app/kent_mpi.cpp: MPI coordinator that launches kent across multiple nodes
• scripts/: database setup, classification, abundance estimation, taxonomy update, and download helpers
• config/cluster.conf.example: sanitized template for cluster execution

Requirements

• Linux with a C++11-capable compiler and GNU Make
• NVIDIA CUDA Toolkit for the full cuCLARK build
• OpenMPI for kent-mpi

Full replication requires a CUDA-enabled system. Building kent alone without CUDA is useful for orchestration development, but it is not a substitute for the full GPU workflow.

Quick Start

./install.sh
./bin/kent --verify

Then configure a database and run classification:

./bin/kent -d /path/to/database
./bin/kent -c -O /path/to/reads.fastq -R results/sample.csv
./bin/kent -a /path/to/database results/sample.csv
./bin/kent -r

Single-Node Workflow

./bin/kent -h
./bin/kent --verify
./bin/kent -d /path/to/database
./bin/kent -c -O /path/to/reads.fastq -R results/sample.csv
./bin/kent -a /path/to/database results/sample.csv
./bin/kent -m results/run1.csv results/run2.csv -o combined.csv
./bin/kent -r

Notes:

• kent -d must be run before classification so that scripts/set_targets.sh can generate scripts/.settings.
• The shell scripts in scripts/ expect to run from the scripts/ directory. kent and kent-mpi handle that automatically.
• Generated results go to results/; logs go to logs/.

MPI Workflow

1. Copy the cluster template:

cp config/cluster.conf.example config/cluster.conf

2. Edit config/cluster.conf for your environment.

3. Run preflight checks, then launch:

./bin/kent-mpi -c config/cluster.conf -p
./bin/kent-mpi -c config/cluster.conf

The MPI configuration file uses INI syntax with [section] headers and key = value pairs.

Repository Layout

See STRUCTURE.md for an accurate file-by-file overview of the public tree.

License

This repository is distributed under the GNU General Public License, version 3 or later. See LICENSE.