Jaeger models

Model weights for Jaeger, a homology-free deep-learning tool to detect bacteriophage and eukaryotic virus sequences in metagenomic assemblies.

Repository layout

Each model lives in its own top-level directory:

<model_name>/
├── model/
│   ├── <model_name>_graph/              # SavedModel for inference
│   ├── <model_name>_embedding_graph/    # SavedModel with embedding output
│   ├── <model_name>.weights.h5          # Keras weights (for jaeger interpret)
│   ├── <model_name>_project.yaml        # build recipe
│   ├── <model_name>_classes.yaml        # class labels
│   └── <model_name>_refine.yaml         # calibration for --refine
└── taxonomy_db/                         # (some models) FAISS taxonomy database
    ├── genomes.faiss
    ├── index2tax.npz
    └── tax/

Available models

Model	Params	Classes	Taxonomy db
jaeger_448931d3_1.3M_fragment	1.3M	bacteria, phage, eukarya, archaea, plasmid, virus	yes (ICTV, FAISS)
jaeger_91c5575c_1.3M_fragment	1.3M	six-class	no
jaeger_848498a0_1.2M_fragment	1.2M	six-class	no

Usage

Install Jaeger, then download a model:

pip install jaeger-bio[cpu]        # or [gpu]
jaeger predict -i contigs.fasta -o results -m jaeger_448931d3_1.3M_fragment

To download directly from the Hub instead of the built-in catalog:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="Yasas1994/Jaeger",
    allow_patterns="jaeger_448931d3_1.3M_fragment/**",
    local_dir="~/jaeger_models",
)

Then register the path with jaeger register-models --path ~/jaeger_models.

Citation

If you use Jaeger in your work, please cite:

• Jaeger: an accurate and fast deep-learning tool to detect bacteriophage sequences Yasas Wijesekara, Ling-Yi Wu, Rick Beeloo, Piotr Rozwalak, Ernestina Hauptfeld, Swapnil P. Doijad, Bas E. Dutilh, Lars Kaderali bioRxiv, 2024.09.24.612722

Code DOI: https://doi.org/10.5281/zenodo.20534106