Jaeger : an accurate and fast deep-learning tool to detect bacteriophage sequences

• Installation
  • Quick reference
  • One-liner (recommended)
  • Bioconda
  • PyPI
  • Git
  • Apptainer
  • Troubleshooting
    • Jaeger fails to detect the GPU
• Running Jaeger
  • Table of contents
  • Basic prediction
    • Common options
  • Choosing a model
  • Batch size and memory
  • Output files
  • Understanding the output table
    • Filtering tips
  • Prophage extraction
  • Command-line reference
    • Main commands
    • jaeger predict full help
    • Utility commands (jaeger utils)
  • Python integration
• Performance optimizations
  • Quick comparison
  • 1. Mixed precision
    • When to use
    • Caveats
  • 2. XLA JIT compilation
    • When to use
    • Caveats
  • 3. ONNX Runtime
    • 3.1 Install dependencies
    • 3.2 Convert the model
    • 3.3 Run inference
    • Provider selection
    • When to use
  • 4. TFLite quantization
    • 4.1 Quantize the model
    • 4.2 Run inference
    • When to use
    • Caveats
  • 5. ONNX INT8 quantization
    • 5.1 Install dependencies
    • 5.2 Convert and quantize
    • 5.3 Run inference
    • When to use
    • Caveats
  • 6. TensorRT (TF-TRT)
    • 6.1 Use an NGC container
    • 6.2 Easier alternative
    • When to use
  • Environment configuration
    • For ONNX Runtime + TensorRT
    • Verify providers
  • Choosing an optimization
• Training and Fine-Tuning Models
  • Table of contents
  • Overview
  • Training workflow
  • Preparing training data
    • Step 1: Collect reference sequences
    • Step 2: Generate training fragments
    • Step 3: Simulate metagenome fragments
    • Step 4: Generate OOD data for reliability training
  • Configuration file
    • Minimal example
    • Key configuration sections
  • Running training
    • From scratch
    • With mixed precision (faster on modern GPUs)
    • Resume from checkpoint
    • Save model without training
  • Fine-tuning
    • Freeze the representation learner, train only heads
    • Train only the classification head
    • Train only the reliability head
    • Tips for fine-tuning
  • Self-supervised pretraining
  • Creating ensembles
  • Command reference
    • jaeger train
    • jaeger register-models
  • Data augmentation tips
    • Sequence masking
    • Sequence mutation
• Releasing Jaeger
  • Versioning scheme
  • Prerequisites
  • Step 1: Bump the version
    • Option A — Auto-bump (recommended)
    • Option B — Explicit version
    • What the script updates
  • Step 2: Commit and tag
  • Step 3: Push to trigger CI
  • Step 4: What happens automatically
    • 1. publish-to-pypi.yaml
    • 2. release.yaml
    • 3. bioconda-update.yaml
  • Step 5: Verify the release
    • PyPI
    • Bioconda
    • GitHub Release
  • OpenID Connect (OIDC) setup
    • 1. Create a PyPI project (or claim an existing one)
    • 2. Add a Trusted Publisher
    • 3. Add TestPyPI publisher (optional but recommended)
    • 4. How it works in the workflow
    • 5. Troubleshooting
  • Manual fallback
    • Manual PyPI publish
    • Manual Bioconda update
    • Manual GitHub Release
  • Repository guard