Splice Neoepitope Pipeline – Reproducible Snakemake Workflow

Author and Maintainer (Open Source) · Mar 2026 – present

Modernised, open-source reimplementation of the 2015 SNU splice-junction–based neoepitope discovery work as a reproducible Snakemake pipeline with optional structural validation.

Technologies

Snakemake 8.xCondaPython 3.11HISAT2 / STAROptiTypeMHCflurry 2.xTCRdock / AlphaFold v2bedtoolsGoogle Cloud GPUMol*

Contributions

Re-engineered the original SNU pipeline as a fully reproducible Snakemake 8.x workflow with per-rule Conda environments — installable without manual dependency management; run locally and on GCP GPU VMs, and executor-portable to a SLURM cluster by design (per-rule resources + Conda) — a config swap, not a rewrite.
Implemented tumor-vs-matched-normal junction filtering against GENCODE annotation, classifying junctions as annotated, normal_shared, or tumor_exclusive.
Integrated patient-specific HLA typing (OptiType) and MHC-I peptide presentation prediction (MHCflurry 2.x Class1PresentationPredictor) for end-to-end neoepitope candidate ranking.
Added optional TCR-pMHC ternary complex structural validation via TCRdock (AlphaFold v2 backend) on Google Cloud GPU; predicted complexes rendered as interactive Mol* 3D viewer in the report.
MIT-licensed, public on GitHub (Jin-HoMLee/splice-neoepitope-pipeline).

Outcome

Snakemake pipeline: end-to-end neoepitope discovery from RNA-Seq FASTQ to computationally modeled TCR-pMHC complex structure — patient-specific HLA alleles, tumor-exclusive junctions.