empathi 1.0.4

An embedding-based phage protein annotation tool by hierarchical assignment

Empathi
Embedding-based Phage Protein Annotation Tool by Hierarchical Assignment

Table of Contents

1. About the Project
2. Getting Started
   • Prerequisites
   • Installation
3. Usage details

About the Project

Empathi is a tool for the prediction of bacteriophage protein functions. It utilizes the highly informative ProtT5 protein embeddings to make predictions. In addition, new functional groups were defined to be better suited for machine-learning than the often-overlapping PHROG categories.

A preprint is available here.

Getting Started

Empathi has been packaged in PyPI and as an Apptainer container for ease of use.
The source code can also be downloaded from HuggingFace.

Prerequisites

The full list of dependencies and versions can be found in requirements.txt.

Either git-lfs or Apptainer will be required. See instructions below.

Other dependencies are taken care of by pip and Apptainer.

python/3.11.5
joblib==1.2.0
numpy==1.26.4
pandas==2.2.1
torch==2.3.0
scipy==1.13.1
scikit-learn==1.5.0
transformers==4.43.1
sentencepiece==0.2.0

Installation

There are three ways of installing Empathi: through PyPI, as an Apptainer container or as source code.

1. PIP

First, create a virtual environment in python 3.11.5.

conda create -n empathi_env python=3.11.5
conda activate empathi_env

Download models for Empathi. You will need git-lfs: for WSL or linux use sudo apt-get install git-lfs, for windows either use git bash or get it from here. Then:

git lfs install
git clone https://huggingface.co/AlexandreBoulay/empathi
export PATH="/path/to/empathi/models:$PATH"

Install dependencies:

pip install empathi

Usage

empathi input_file name

2. Apptainer

Download Apptainer or singularity. On windows, this will require a virtual machine. WSL works well.

Fetch Empathi from Sylabs:

apptainer pull empathi.sif library://alexandreboulay/empathi/empathi

Usage

apptainer run empathi.sif path/to/input_file name

3. From source code

First, create a virtual environment in python 3.11.5.

conda create -n empathi_env python=3.11.5
conda activate empathi_env

Clone the repo. You will need git-lfs: for WSL or linux use sudo apt-get install git-lfs, for windows either use git bash or get it from here. Then:

git lfs install
git clone https://huggingface.co/AlexandreBoulay/empathi

Install dependencies:

cd empathi
pip install -r requirements.txt

Usage

python src/empathi/empathi.py input_file name

Usage details

A fasta file of protein sequences or a csv file of protein embeddings can be used as input.

Specifying the option --only_embeddings will only compute embeddings. This step is much faster with a GPU. The embeddings file can then be reinputted using the same command (without --only_embeddings) and specifying the new file as input file.

Options:

• input_file: Path to input file containing protein sequencs (.fa*) or protein embeddings (.pkl/.csv).
• name: Name of file you want to save to (wOut extension). Should be different between runs to avoid overwriting files.
• --models_folder: Path to folder containing EmPATHi models. Can be left unspecified if it was added to PATH earlier.
• --only_embeddings: Whether to only calculate embeddings (no functional prediction).
• --output_folder: Path to the output folder. Default is ./empathi_out/.
• --mode: Which types of proteins you want to predict. Accepted arguments are "all", "pvp", "rbp", "lysin", "regulator"...

When launching from python omit the '--' in front of args.

Output format

A csv file with an annotation column regrouping all assigned annotations per protein (separated by "|") and a column per functional category with the confidence associated to each prediction. A confidence >0.5 is used to assign functions, but a user could raise this criteria by filtering the output table.

Ex.

Annotation	PVP	cell wall depolymerase	DNA-associated	...
PVP|cell wall depolymerase	0.98	0.99	0.005	...
DNA-associated	0.01	0.05	0.998	...