PhaNet

PhaNet annotates phage genomes and provides protein function prediction, module analysis, RBP screening and genome classification.

Requirements

• Linux.
• Python 3.10 or later (3.11 recommended).
• Conda or Mamba.

For the Quick start, we recommend 4–8 CPU cores, 32 GB RAM and 150 GB free disk space. An NVIDIA GPU is optional; GPU use requires a working NVIDIA driver.

See the installation guide for dependencies, GPU setup and resource recommendations for other workflows.

Quick start

Annotate the λ (Lambda) phage genome included at examples/lambda-genome.fna. This example runs gene prediction, traditional annotation and PhaNet prediction on CPU. The input is the 48,502 bp genome from ENA J02459.1.

The commands below install and configure GLM mode from the extracted PhaNet source. Choose a writable directory on your data disk; change the path in the first line below to your location:

export PHANET_DATA="/data/phanet-data"
mkdir -p "$PHANET_DATA"

Download the phanet resource folder from Google Drive and extract all ZIP parts into that directory, keeping the downloaded folder name. The bundle should be at $PHANET_DATA/phanet/, with RESOURCE_MANIFEST.json directly inside it.

For Module, GO, VF, RBP and taxonomy models, see model installation.

1. Install and configure once

From the source directory containing pyproject.toml and examples/, copy:

conda create -n phanet python=3.11 pip -y
conda activate phanet
python -m pip install 'torch>=2.1,<3' --index-url https://download.pytorch.org/whl/cpu
python -m pip install '.[esm]'

phanet setup --install-tools \
  --eggnog-data "$PHANET_DATA/eggnog/5.0.2" \
  --config "$PHANET_DATA/traditional.json"
phanet setup-resources --method glm \
  --bundle "$PHANET_DATA/phanet" \
  --external "$PHANET_DATA/external" \
  --out "$PHANET_DATA/config-glm" \
  --config "$PHANET_DATA/traditional.json"

The setup commands install the annotation tools, download eggNOG and ESM2, and generate configuration files. For other installation options, see the installation guide.

2. Run the example

In the same source directory:

source "$PHANET_DATA/config-glm/activate.sh"
phanet annotate --input examples/lambda-genome.fna --out lambda-analysis/annotation \
  --method glm --device cpu --threads 4 --batch-size 1
cat lambda-analysis/annotation/results/summary.json

With the tools and database installed above, the example produces 62 proteins: 53 have traditional annotations, 7 receive additional PhaNet labels and 2 remain unassigned. Open lambda-analysis/annotation/results/annotations.tsv for the annotations. The same directory contains proteins.faa, annotations.gff3 and summary.json.

For your own data, replace examples/lambda-genome.fna with your nucleotide FASTA or FASTA.gz file and lambda-analysis/annotation with a new output directory. In a new shell, set PHANET_DATA to the same location, run conda activate phanet and source "$PHANET_DATA/config-glm/activate.sh" before annotation. To continue the same run, add --resume.

Optional workflows

Task	How to use it
Run on GPU	Install CUDA PyTorch as described in installation, then use --device cuda
Add network annotation or use combined mode	Install the network dependencies and resources using installation steps 1 and 4; then select --method network or --method combined
Use existing tools, change data locations or pass explicit paths	Configuration and data paths
Run gene prediction and traditional annotation only	Preparation
Build a network from your own references or prepare GLM training data	Custom data
Train, fine-tune, evaluate or export a model	Training
Predict modules, GO terms or VF labels	Downstream prediction
Compare and group predicted modules	Module analysis
Predict RBP, classify genomes or summarize domains	RBP and classification

See reading and exporting results, running and recovery, troubleshooting and more examples for common tasks. Run phanet --help or phanet <command> --help for command options.

Python interfaces and code structure

Use the Python API for function calls, return values and output paths. The architecture overview shows the main modules and processing flow.

License

PhaNet uses the MIT License. Database and model sources are listed in resource sources and third-party notices.