tsarina

Tsarina selects shared cancer immunotherapy targets for an individual patient or a population HLA panel.

It starts with reusable cancer-testis antigen (CTA), oncogenic-virus, and recurrent-mutation targets. It then combines tumor context, public immunopeptidomics evidence, healthy-tissue safety evidence, and predicted HLA presentation to produce ranked peptide-MHC (pMHC) candidates.

Workflows

Goal	Entry point	Result
Prioritize targets for one patient	`tsarina personalize` or `personalized_targets()`	Ranked CTA, viral, and mutant pMHCs for that patient's HLA type and tumor
Design an off-the-shelf CTA panel	`tsarina panel`	CTA × HLA matrix with evidence tiers and population-coverage estimates
Assemble a shared CTA vaccine	`tsarina vaccine`	Mortality-ranked proteoforms, sequence funnel, junction audit and DNA/RNA construct
Inspect public peptide observations	`tsarina hits`	Cancer, healthy-tissue, and restriction evidence for specified peptides

See: documentation — inputs, dataset setup and workflow details.

Install

pip install tsarina

Install the optional peptide-generation and partitioning dependencies for full functionality:

pip install "tsarina[all]"

Quick start

Prioritize targets for a patient:

tsarina personalize \
  --hla 'HLA-A*02:01,HLA-A*24:02,HLA-B*07:02' \
  --cta 'MAGEA4=142.5,PRAME=87.3' \
  --mutations "KRAS G12D" \
  --viruses hpv16 \
  --output patient-targets.csv

Build the default global CTA × HLA panel:

tsarina panel --format long --output panel.csv

Both workflows can use registered IEDB or CEDAR ligand exports as public immunopeptidomics evidence. See Data and evidence for setup and interpretation.

Target categories

Category	Candidate source	Tumor-specific context
CTA	The canonical oncoref CTA set	Tumor RNA expression and reproductive-tissue restriction
Viral	Proteomes from nine oncogenic viruses	Virus detected in the tumor
Mutant	Nineteen recurrent hotspots across seven driver genes	Matching mutation detected in the tumor

Oncoref is the single authority for CTA membership, aliases, HPA restriction calls, and proteoform groups. Tsarina adds downstream evidence and scoring without maintaining a second CTA definition library. The exact boundary is documented in CTA ownership and downstream evidence.

How ranking works

Target selection follows five stages:

choose candidates from the relevant shared-target sets;
enforce tumor context and exclude non-target human peptide matches;
annotate public cancer and healthy-tissue MS observations;
predict presentation by the requested HLA alleles; and
rank pMHCs with explicit evidence and safety provenance.

Public MS evidence strengthens a candidate but does not redefine CTA membership. Healthy, direct-ex-vivo observations outside reproductive tissues and thymus are treated as safety evidence.

Documentation

Documentation guide — workflow selection and shared concepts
Personalized target selection — patient inputs, prioritization, and output schema
CTA panel design — automatic selection, evidence tiers, HLA panels, and coverage
CTA vaccine design — mortality-weighted p95 selection, strict/loose specificity, MS-supported stretches, construct search and full audit
Canine evidence and DLA design — frozen dog inputs, independent-dog RNA prevalence, exact MS support and exploratory DLA constructs
Data and evidence — data registry, observation classification, scoring, and naming
CTA ownership and downstream evidence — oncoref/Tsarina responsibilities and maintenance

Development

./develop.sh
./format.sh
./lint.sh
./test.sh

develop.sh uses the active virtualenv (or creates .venv) and requires pip >=23.0. It resolves Tsarina and present sibling checkouts together, checks the proposed versions against installed consumers before changing packages, and runs pip check after installation. If a checkout is incompatible, update that checkout or use a separate virtualenv; the refused install leaves the installed packages unchanged. Missing siblings use their released packages.

Real MHCflurry model integration tests run in a separate CI job. To run them locally (Python 3.10+):

pip install -e ".[test-mhcflurry]"
mhcflurry downloads fetch models_class1_presentation
pytest tests/test_mhcflurry_integration.py --run-mhcflurry

These tests check the downloaded presentation bundle, including its affinity calibration, through Tsarina's scoring interface. Enabling them requires the dependency and models; missing prerequisites fail rather than skip. Use ./test.sh --run-mhcflurry to include them in the full suite. ./deploy.sh requires this extra and model bundle and always runs these tests.

Metadata

Release files for tsarina 1.35.1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for tsarina 1.35.1
File	Size	Uploaded
tsarina-1.35.1.tar.gz	2.1 MB	Oct 9, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for tsarina 1.35.1
File	Interpreter	ABI	Platform	Reset
tsarina-1.35.1-py3-none-any.whl 675.7 kB Oct 9, 2026	Python 3	none	any	Details

Total release size: 2.8 MB

Release files / tsarina-1.35.1.tar.gz

Download URL	tsarina-1.35.1.tar.gz
Size	2.1 MB
Tags	Source
SHA-256 checksum How to use checksums	`795628e077c7e28bfde6060a86a89ef9f071ad205b3e478f81f8407f506f749a`
BLAKE2b-256 checksum How to use checksums	`96997acc1d50209d012fb8d674357bd650cfd2bb3db4100def1883e5c6ebddb1`
Upload date	Oct 9, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.6`

Release files / tsarina-1.35.1-py3-none-any.whl

Download URL	tsarina-1.35.1-py3-none-any.whl
Size	675.7 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`a1c6c62e7e35c55f59078461aae6f0beb5c26b9b593e1b6904ca2259d8f88c56`
BLAKE2b-256 checksum How to use checksums	`a81cbe8c60c89a0517aea5014caa0b25bed59a87649100d9f5f66a3fcafaf36b`
Upload date	Oct 9, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.12.6`