Ruspladder

Ruspladder is an independent Rust implementation of the non-visual analysis in SplAdder v3.1.1. It takes gene annotation and RNA-seq alignments, builds and augments splice graphs, detects alternative splicing events, quantifies their support and PSI, and tests differences between conditions. It implements prep, build and differential test, including exon skipping, intron retention, alternative 3′/5′ splice sites, multiple-exon skipping and mutually exclusive exons.

The scientific methods originate in SplAdder. Ruspladder is maintained by boman-ng; it is not an official release of SplAdder, ratschlab or OpenGene. Its focus is preserving tested scientific behavior while reducing runtime and memory use through Rust, bounded data processing and multithreading. See compatibility and differences before substituting it in an existing workflow.

Install the Linux binary

Download ruspladder-v0.1.0-linux-amd64.tar.gz and its .sha256 file from Releases. The release is public; no GitHub account is required:

curl -fLO https://github.com/boman-ng/ruspladder/releases/download/v0.1.0/ruspladder-v0.1.0-linux-amd64.tar.gz
curl -fLO https://github.com/boman-ng/ruspladder/releases/download/v0.1.0/ruspladder-v0.1.0-linux-amd64.tar.gz.sha256
sha256sum -c ruspladder-v0.1.0-linux-amd64.tar.gz.sha256
tar -xzf ruspladder-v0.1.0-linux-amd64.tar.gz
cd ruspladder-v0.1.0-linux-amd64
export OPENBLAS_NUM_THREADS=1
./ruspladder --version

Requires Linux amd64 (x86_64), glibc 2.28 or newer, including RHEL/AlmaLinux 8, Debian 10+ and Ubuntu 20.04+. Keep the executable and adjacent lib/ directory together; the whole directory can be moved. Python, Rust, a compiler and a system BLAS/HDF5 installation are unnecessary at runtime. Alpine/musl and ARM are outside this release's supported platform. OPENBLAS_NUM_THREADS=1 prevents the bundled BLAS loader from creating an idle thread pool; --parallel controls analysis workers.

Quick start

Use coordinate-sorted, indexed BAM files and a matching GTF/GFF3 annotation. For CRAM, also supply --reference genome.fa.

./ruspladder build -a annotation.gtf \
  -b control1.bam,control2.bam,treated1.bam,treated2.bam \
  -o results --parallel 4 --readlen 150 --output-txt

./ruspladder test -o results \
  -a control1,control2 -b treated1,treated2 --parallel 4 --readlen 150

Sample names for test correspond to the alignment basenames without extensions. Set --readlen to your data's read length. Public results include graph and gene expression counts, event counts/PSI in HDF5, optional event text files, and non-visual differential-testing TSV files. Use ./ruspladder COMMAND --help for parameters. Internal Rust caches cannot be exchanged with Python pickle caches.

For reusable sparse alignment summaries:

./ruspladder prep -a annotation.gtf -b sample.bam --sparse-bam --parallel 4
./ruspladder build -a annotation.gtf -b sample.bam -o results-sparse \
  --sparse-bam --parallel 4

For GTFs reusing a gene ID across reference sequences or strands, explicit --annotation-mode locus writes normalized .locus.gtf and .loci.tsv companions beside the annotation. It preserves coordinates and records original identities. This changes affected gene models: compare Python and Rust on the same normalized GTF. Default --annotation-mode spladder retains upstream import behavior. Use separate result directories for the two modes. Annotation files and their caches are treated as immutable; after an edit, use a new path or remove the generated companions. See annotation details.

Resources and verification

The resource baseline is 4 logical CPUs / 4 GiB RAM, with a hard memory limit and swap disabled. Commands accept 1–64 threads. Memory requirements depend on annotation spans, sample count and event density; 4 GiB is the benchmark configuration, not a cap enforced by the executable or a promise for every input.

make test replays the six non-visual scenarios in SplAdder's public tests/test_end_to_end.py: positive/negative strand, merged/single analysis, and 20-sample BAM/CRAM build plus differential testing. Both programs run from fresh annotations; graphs, events, HDF5 values and text/TSV results are compared. CI tests the release package and its relocation. COMPATIBILITY.md defines comparison precision and the limits of this evidence.

Build and test from source

Build/reference caches and data default to ~/data/ruspladder; override with RUSPLADDER_WORK_ROOT. The supported build uses the pinned Linux toolchain:

export RUSPLADDER_WORK_ROOT="$HOME/data/ruspladder"
mkdir -p "$RUSPLADDER_WORK_ROOT"
docker build -f Dockerfile.build -t ruspladder-build:0.1.0 .
docker run --rm --cpus 4 --memory 4g --memory-swap 4g \
  --user "$(id -u):$(id -g)" \
  -v "$PWD:/src" -v "$RUSPLADDER_WORK_ROOT:/work" \
  ruspladder-build:0.1.0 bash scripts/ci.sh

This builds, checks, packages and compares public data. The runtime archive, corresponding native-source archive, checksums and verification JSON are written to $RUSPLADDER_WORK_ROOT/dist/. See redistribution materials. Use a fresh work directory when rerunning this complete packaging/CI command. Python is used only for dependency extraction, packaging and reference comparisons. NUMERICS.md describes why the native numerical dependencies are pinned.

With Rust ≥1.89, C/C++17, Clang/libclang, CMake ≥3.26, zlib/bzip2 development headers and Python ≥3.11 installed, make build builds locally. The maintenance scripts use Python 3.11 standard-library hashing and TOML APIs. To run comparisons, install uv, run bash scripts/bootstrap.sh, then make test. The broader comparison suite is available through bash scripts/check.sh after python3 scripts/fetch_fixtures.py --work-root "$RUSPLADDER_WORK_ROOT".

Attribution, citation and license

Thank you to André Kahles, Cheng Soon Ong, Yi Zhong and Gunnar Rätsch for the SplAdder methods and publication, and to the SplAdder authors and contributors for making their implementation and test data available. Ruspladder derives its algorithms from SplAdder v3.1.1, commit 65ceec839b9ff0cf96703c1605ee43667662f410. It also reuses work from NumPy, SciPy, statsmodels, OpenBLAS/LAPACK, HTSlib, HDF5 and x86-simd-sort; source notices and provenance are retained in licenses/ and vendor/.

When using Ruspladder in research, cite the original method and record the Ruspladder version:

Kahles A, Ong CS, Zhong Y, Rätsch G. SplAdder: identification, quantification and testing of alternative splicing events from RNA-Seq data. Bioinformatics 32(12), 1840–1847 (2016). doi:10.1093/bioinformatics/btw076. Machine-readable citation metadata is in CITATION.cff.

Ruspladder inherits SplAdder's BSD 3-Clause license, retaining the original copyright and adding attribution for the Rust implementation. See LICENSE. Third-party components retain their own licenses; binary archives include their notices. Upstream attribution does not imply endorsement.

Contributing and maintenance

See CONTRIBUTING.md for development checks and scientific compatibility requirements, and SECURITY.md for private reports.