Bridge

Bridge is a standalone Python package for applying the Bridge model to predict Acute Leukemia class to new RNA-seq count matrices.

The inference bundle (bridge_inference.bundle) is not distributed publicly in this repository. If you want to run the package, please email us (the project maintainers) to request bundle access.

pip install -e ./Bridge

Conda package (published):

conda install -c eonurk -c conda-forge -c pytorch bridge

Package page: https://anaconda.org/eonurk/bridge

from bridge import BridgePredictor

predictor = BridgePredictor(
    bundle="path/to/bridge_inference.bundle",
)

result = predictor.predict_rna("path/to/rna_count_matrix.csv")
print(result.combined.head())
predictor.close()

result.combined includes predictions, per-class probabilities, and RNA latents. The bundle is provided by the model team and contains all required inference artifacts.

After install, the package exposes a bridge command:

bridge --help

Predict labels from an RNA matrix:

bridge predict-rna \
  --bundle path/to/bridge_inference.bundle \
  path/to/rna_count_matrix.csv \
  --combined-out outputs/predictions.csv

Export RNA latents:

bridge encode-rna \
  --bundle path/to/bridge_inference.bundle \
  path/to/rna_count_matrix.csv \
  --output outputs/rna_latents.parquet

Export methylation latents:

bridge encode-methylation \
  --bundle path/to/bridge_inference.bundle \
  path/to/methylation_matrix.csv \
  --output outputs/meth_latents.csv

If you are not using a bundle, pass --checkpoint (plus --metadata if needed) and for predict-rna also pass --classifier.

For existing pipelines, avoid storing large bundles in your repo by passing a URL and cache dir:

bridge predict-rna \
  --bundle-url https://example.org/path/bridge_inference.bundle \
  --bundle-cache-dir /shared/cache/bridge \
  --bundle-sha256 <expected_sha256> \
  input/rna_counts.csv \
  --combined-out output/predictions.csv

The bundle is downloaded once and reused from cache in later runs.

All model-source options support environment variable fallbacks:

• BRIDGE_BUNDLE, BRIDGE_BUNDLE_URL, BRIDGE_BUNDLE_CACHE_DIR, BRIDGE_BUNDLE_SHA256
• BRIDGE_CHECKPOINT, BRIDGE_METADATA, BRIDGE_CLASSIFIER

• License: CC-BY-NC-4.0
• Non-commercial use only.
• For any commercial use (including commercial research, product integration, or deployment), you must email us to request a separate commercial-use license.
• See LICENSE for details.

The package auto-detects whether features are in rows or columns and aligns input features to the training metadata.

A drop-in Snakemake example is included at:

• Bridge/examples/snakemake/Snakefile
• Bridge/examples/snakemake/rules/bridge.smk
• Bridge/examples/snakemake/config.yaml
• Bridge/examples/snakemake/README.md

It is designed for private bundle delivery via environment variables (BRIDGE_BUNDLE or BRIDGE_BUNDLE_URL) and keeps bundle paths/URLs out of committed config.

Install test dependencies and run:

cd Bridge
pip install -e ".[test]"
PYTHONPATH=src pytest -q