Notebooks

Worked examples as marimo notebooks — plain Python files with @app.cell decorators, so they diff and review like source rather than like JSON, and they run three ways:

pip install 'mirpy-lib[examples]'
marimo edit examples/signature_pipeline.py    # interactive
marimo run  examples/signature_pipeline.py    # read-only app
python      examples/signature_pipeline.py    # plain script; prints, no UI

Each one bootstraps its own data from Hugging Face and caches it under examples/.data/, so a fresh pip install is enough — no local paths and no pre-staged files. Drop a copy under ./data_dump/ (gitignored) and it is used instead of downloading.

Start here

Signature pipeline

A folder of AIRR TSVs to one table joined with your metadata, in one command --- then reading the result by channel. Runs on 1,764 SRA samples from isalgo/airr_benchmark. Read this one first.

Quickstart

Clonotypes to vectors: the prototype embedding, what the coordinates mean, and the distance approximation it rests on.

Signature internals

Why the geometry half is transformed differently from the statistics half, and how many principal components survive a group-disjoint refit.

Going further

Feature vectors

The repertoire-level embedding in full: bases, weights, and what varies between donors.

Density and enrichment

Neighbourhood density, TCRnet-style enrichment, and reading out what drives a signal.

Trajectories and twins

Tracking a repertoire through time, and comparing donors who should look alike.

Theory

The results T1-T7 the library rests on, each with the measurement that supports it.