Leech¶

Learning Enhanced Electrical Classifiers from Hanopore signals

Leech trains classifiers for motif-centered tasks on Oxford Nanopore signal data -- modified-base calling, or any other label attached to a sequence motif. It extracts dwell time features from move tables (the BAM mv tag) and feeds them alongside raw signal and sequence context into a multi-branch neural network, giving it information that signal-only tools like Remora discard.

Development and validation used tRNA aminoacylation ("charging") state and amino acid identity classification from direct RNA sequencing as the driving example -- it shows up throughout the docs and the bundled Snakemake pipeline -- but nothing in the feature extraction, model, or CLI is specific to that assay: motifs, labels, and classes are all user-defined.

Install¶

uv add "leech[rust]"     # or: pip install "leech[rust]"

Released on PyPI as leech plus leech-core, the optional Rust accelerator pulled by the rust extra. See Installation for the from-source path and platform notes.

Workflow¶

graph LR
    A["leech data prepare"] --> B["leech model train"]
    B --> C["leech eval test"]
    C --> D["leech predict"]

1. Prepare -- extract signal, sequence, and dwell features from POD5 + BAM files
2. Train -- fit a multi-branch neural network on the extracted features
3. Test -- evaluate on held-out data (accuracy, AUC, confusion matrix)
4. Predict -- apply the model to new reads and write predictions to BAM

Documentation¶

Citation¶

If you use leech, please cite:

• This work (publication pending)
• Remora (underlying training framework)

License¶

MIT License -- see LICENSE for details.