v2.0.3

What's new

Search

• --class {I,II,all}: restrict the reference to MHC class I, II, or all (default all) for faster, more specific searches.
• --always-top-n: return each cluster's top-N matches regardless of threshold; sub-threshold hits are flagged below cutoff in the report. Opt-in, off by default.

Report (standalone HTML)

• Redesigned layout: per-cluster-solution motif carousel with a solution dropdown (All by default) and X-of-N navigation.
• Paginated, searchable Top-N results table; left table-of-contents with scroll-spy.
• Info popovers on each section/plot; static PCC colourbar with a live threshold marker; network group colour legend.
• Canonical allele names that match the embedded reference logo titles.
• Browser favicon; citation + repository links; graceful placeholders for missing logos.

Documentation

• New MkDocs Material site with API reference auto-generated from docstrings, plus a standalone example report.
• Pearson-correlation method and formula documented in the engine docstrings.

Install: pip install -U mhc-tp

v2.0.2

Packaging polish — a richer, more professional PyPI project page.

Changes

• Added project links shown on PyPI: Homepage, Documentation (https://purcelllab.github.io/MHC-TP/), Repository, Issues, and Changelog.
• Added Trove classifiers (Bio-Informatics, Python 3.9–3.11, Production/Stable, Console) and keywords for discoverability.
• Declared a maintainer entry.

No functional changes from v2.0.1.

Install

pip install mhc-tp
mhc-tp fetch -s human
mhc-tp search <gibbscluster_dir> -s human -o results/

v2.0.1

The project's PyPI package is now mhc-tp (renamed from hla-pepclust). Functionally identical to v2.0.0.

Install

pip install mhc-tp
mhc-tp fetch -s human
mhc-tp search <gibbscluster_dir> -s human -o results/

• Distribution: mhc-tp · command: mhc-tp · import: import mhc_tp
• The previous hla-pepclust 2.0.0 on PyPI has been yanked; please use mhc-tp.

See the v2.0.0 notes for the full feature list (modern packaging, numba engine, class I+II human/mouse reference fetched from the refdata-2025 release, standalone HTML report).

v2.0.1rc1

Release candidate for the mhc-tp rename (formerly hla-pepclust). Published to TestPyPI.

pip install -i https://test.pypi.org/simple/ mhc-tp==2.0.1rc1

v2.0.0

A ground-up modernization of HLA-PepClust into a clean, pip install-able tool, focused on its core science: cluster-to-allele motif matching, shipped as a fast CLI with a standalone interactive HTML report.

Install

pip install hla-pepclust
clust-search fetch -s human        # download the reference (human | mouse | all)
clust-search search <gibbscluster_dir> -s human -o results/

Highlights

• Standard packaging — src/ layout, pyproject.toml (hatchling), console entry point clust-search, published to PyPI.
• Class I + II, human & mouse — one compressed Parquet reference per species, built from NetMHCpan-4.2 / NetMHCIIpan-4.3.
• Fetch-on-demand reference — clust-search fetch pulls the reference from the refdata-2025 release and verifies it by SHA-256 (nothing huge committed to git).
• Fast numba engine — JIT correlation search with bounded thread pools, so it stays a good citizen on shared servers.
• Standalone HTML report — responsive, logos embedded as data-URIs, opens with a double-click; Seq2Logo reference motifs included.
• Rich CLI — banner, coloured staged logs, results table.
• Publication-ready names — Immunolyser-style allele labels (HLA_B39124, H2_Kb) in the report and CSV; the CSV keeps a raw formatted key for downstream joins.

Commands

command	purpose
search	correlate GibbsCluster matrices against the reference -> CSV + HTML report
fetch	download the prebuilt reference Parquet(s)
export-logos	dump embedded reference logos to PNG (all, or a chosen allotype list)
build-ref / build-db	(dev) rebuild the reference from NetMHCpan packs

Breaking changes

• The Flask/Celery web application has been removed — this is now CLI + HTML report only.
• The reference is no longer bundled; run clust-search fetch once after install.
• CLI invocation is now clust-search search ... (see README / --help).

Notes

• Reference data lives on the Reference data 2025 release and is SHA-256 verified.
• Reference motifs are derived from NetMHCpan-4.2 / NetMHCIIpan-4.3 and rendered with Seq2Logo-2.1 (DTU Health Tech) for academic/research use.
• Tested on real human and mouse immunopeptidomics datasets; 226 automated tests, CI on Linux.

v2.0.0rc1

Release candidate for v2.0.0 — published to TestPyPI for verification.

pip install -i https://test.pypi.org/simple/ hla-pepclust==2.0.0rc1

Full notes will accompany the final v2.0.0 release.

Reference data 2025

Prebuilt HLA/MHC reference motifs for clust-search (human + mouse, class I + II).

Downloaded automatically by clust-search fetch. SHA-256 sums are verified against reference_manifest.tsv.

Provenance / attribution

• Class I motifs derived from NetMHCpan-4.2 (DTU Health Tech)
• Class II motifs derived from NetMHCIIpan-4.3 (DTU Health Tech)
• Reference sequence logos rendered with Seq2Logo-2.1 (DTU Health Tech)

Derived data redistributed for academic/research use.