tcrtrie 0.2.3

Python bindings for TCR trie search

TCRtrie

TCRtrie is a tool for approximate search in TCR repertoires based on a CDR3 index. It can be used both for searching user-provided repertoires and for searching the VDJdb database.

The library supports two search modes:

• edit-distance-based search with bounded substitutions, insertions, and deletions;
• matrix-based search where substitutions are scored using an amino acid substitution matrix.

Main objects

VDJdb

VDJdb is a lazy object. It is initialized on first use and is not callable.

Trie

Trie is the low-level class for building an index from your own AIRR-like TSV file or from in-memory sequence arrays.

Examples

Search in VDJdb with edit distance

from tcrtrie import VDJdb

df = VDJdb.search(  # returns pandas.DataFrame
    query="CASSEGTDGYTF",
    maxSubstitution=2,
    maxInsertion=1,
    maxDeletion=1,
    maxEdits=2,
    vGeneFilter="TRBV19*01",
    jGeneFilter="TRBJ1-2*01",
    numThreads=8,
    detailed=True,
)

df

Batch search in VDJdb

from tcrtrie import VDJdb

VDJdb.searchForAll(
    queries=["CASSEGTDGYTF", "CAISTGDSNQPQHF"],
    maxSubstitution=2,
    maxInsertion=1,
    maxDeletion=1,
    maxEdits=2,
    vGeneFilters=["TRBV19*01", "TRBV6-6*01"],
    jGeneFilters=["TRBJ1-2*01", "TRBJ1-5*01"],
    numThreads=8,
    detailed=True,
)

Search in VDJdb with a substitution matrix

from tcrtrie import VDJdb

VDJdb.searchWithMatrix(
    query="CASSEGTDGYTF",
    maxCost=12,
    detailed=True,
)

Build a trie from your own TSV file

from tcrtrie import Trie

trie = Trie("my_repertoire.tsv")

hits = trie.SearchIndices(
    query="CASSEGTDGYTF",
    maxSubstitution=2,
    maxInsertion=1,
    maxDeletion=1,
    maxEdits=2,
)

hits

Load a custom substitution matrix

from tcrtrie import Trie

trie = Trie("my_repertoire.tsv")
trie.LoadSubstitutionMatrix("my_matrix.txt", delimiter="", gapFactor=1.5)

hits = trie.SearchIndicesWithMatrix(
    query="CASSEGTDGYTF",
    maxCost=12,
)

hits

Installation

Requirements

• Python 3.8+
• pip
• C++17-compatible compiler
• CMake

On Windows, you may need to install Microsoft C++ Build Tools if no suitable compiler is available.

Install from PyPI

pip install tcrtrie

Install directly from GitHub

python -m pip install git+https://github.com/MikePodsytnik/TCRtrie@0.2.0-tcrtriepy

Clone and build locally

git clone --branch TCRtriePy https://github.com/MikePodsytnik/TCRtrie.git
cd TCRtrie
python -m pip install --upgrade pip setuptools wheel scikit-build-core pybind11
python -m pip install .

VDJdb database management

TCRtriePy does not update VDJdb automatically in order to avoid silent changes in scientific results. Database updates are performed explicitly via a command-line tool.

The tcrtrie-vdjdb-update command downloads and installs a selected VDJdb release into the local cache (~/.cache/tcrtrie/vdjdb). The cached version is then used by the VDJdb object in Python.

Install the latest available VDJdb release:

tcrtrie-vdjdb-update

Install a specific VDJdb version (recommended for reproducibility):

tcrtrie-vdjdb-update --tag 2025-12-29

List available VDJdb releases:

tcrtrie-vdjdb-update --list | head -n 10

After updating the database, restart the Python process or Jupyter kernel to ensure the new version is used.

Input data format

Repertoire TSV format

Trie expects an AIRR-like tab-separated file.

Required column:

• junction_aa

Optional columns:

• v_call
• j_call
• __group_id

Minimal example:

junction_aa	v_call	j_call
CASSEGTDGYTF	TRBV19*01	TRBJ1-2*01
CAISTGDSNQPQHF	TRBV6-6*01	TRBJ1-5*01

Notes:

• the file must be tab-separated;
• amino acid sequences are read from junction_aa;
• Sequences must be uppercase and contain only the 20 standard amino acid symbols (ACDEFGHIKLMNPQRSTVWY).
• v_call and j_call are optional, but they are required if you want to use V/J filtering;

Substitution matrix format

The substitution matrix must be a square matrix over amino acid symbols.

Requirements:

• all 20 standard amino acids must be present;
• the matrix may contain either 20 labels or 21 labels if gap (-) is provided explicitly;
• row and column labels must match;
• diagonal values must be strictly greater than every other value in the corresponding row and column.

Whitespace-separated matrices are supported by default. If your matrix uses another separator, pass it via the delimiter argument.

Example:

   A  R  N  D ...
A  4 -1 -2 -2 
R -1  5  0 -2 
N -2  0  6  1 
D -2 -2  1  6 
...

How gap scores are handled

If the matrix already contains -, those values are used directly.

If a gap column is not provided, the gap score for amino acid a is derived as the negative diagonal score, i.e. gap(a) = -score(a, a) * gapFactor. Then the same synthesized value is written to both aa -> - and - -> aa. By default, gapFactor=1.0.

After that, the score matrix is converted into an internal non-negative cost matrix used by matrix-based search.