Sublyme

<h1 align="center">SUBLYME</h1> <div align="center"> <strong>S</strong>oftware for <strong>U</strong>ncovering <strong>B</strong>acteriophage <strong>LY</strong>sins in <strong>ME</strong>tagenomic datasets</div> <br> <!-- TABLE OF CONTENTS --> <details open> <summary>Table of Contents</summary> <ol> <li> <a href="#about-the-project">About the Project</a> </li> <li> <a href="#getting-started">Getting Started</a> <ul> <li><a href="#prerequisites">Prerequisites</a></li> <li><a href="#installation">Installation</a></li> </ul> </li> <li><a href="#usage-details">Usage details</a></li> <li><a href="#output-format">Output format</a></li> <li><a href="#output-format">Citation</a></li> </ol> </details>

About the Project

SUBLYME is a tool to identify bacteriophage lysins. It utilizes the highly informative ProtT5 protein embeddings to make predictions and was trained using the UniProt-derived proteins found in PHALP 1.0.

SUBLYME was then applied to EnVhogDB to create PhaLP 2.0, a metagenomic extension of the previous database.

Getting started

SUBLYME has been packaged in PyPI for ease of use. The source code can be downloaded from GitHub.

Prerequisites

A GPU is recommended to compute embeddings for large datasets.

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

Dependencies are taken care of by pip.

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==1.3.0
transformers==4.43.1
sentencepiece==0.2.0

Installation

First create a virtual environment in python 3.11.5. For example:

conda create -n sublyme_env python=3.11.5
conda activate sublyme_env

From pypi:

pip install sublyme

Usage

sublyme test/input.faa -t 4

SUBLYME accepts as input a multifasta file of protein sequences or a csv file of ProtT5 embeddings.

From apptainer:

Although a little larger (5.8G), the Apptainer container accepts genomic sequences (or protein sequences as well). This allows for a pipeline that launches Prodigal to extract gene and protein sequences, and launches SUBLYME to predict lysins.

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

Fetch SUBLYME from Sylabs:

apptainer pull sublyme.sif library://alexandre_boulay/sublyme/sublyme

Usage

apptainer run sublyme.sif test/input.fa path/to/output_folder {protein|genome} nb_threads [--no-dedup]

Arguments must be specified in the order they appear above, only --no-dedup is optional. The apptainer image accepts either protein or genomic sequences as input. You must specify which. Make sure to specify --no-dedup if you do not want to remove duplicate sequences.

The script always outputs:

• proteins.csv: protein embeddings computed using ProtT5.
• sublyme_predictions.csv: predictions obtained from sublyme.
• sublyme_lysins.faa: protein sequences of predicted lysins.

And if genomic sequences were used as input:

• genes.gff: protein genomic information from Prodigal (position of genes in genome).
• proteins.faa: all protein sequences predicted by Prodigal.
• sublyme_lysin_genes.fna: gene sequences for predicted lysins.

From source:

git clone https://github.com/Rousseau-Team/sublyme.git
cd sublyme
pip install -r requirements.txt

ex. python3 src/sublyme/sublyme.py test/input.faa -t 4 --models_folder src/sublyme/models

SUBLYME accepts as input a multifasta file of protein sequences or a csv file of protein embeddings.

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 sequences (.fa*) or protein embeddings (.csv) that you wish to annotate.
• --threads (-t): Number of threads (default 1).
• --output_folder (-o): Path to the output folder. Default folder is ./outputs/.
• --models_folder (-m): Path to folder containing pretrained models (lysin_miner.pkl, val_endo_clf.pkl). Default is src/sublyme/models.
• --only_embeddings: Whether to only calculate embeddings (no lysin prediction).

Output format

The output consists of a csv file with a column for the final prediction and one column each for probabilities associated to lysins, endolysins and VALs.

Ex. | | pred |lysin|endolysin|VAL | |------|---------------------------|-----|---------|----| |Prot 1| lysin|endolysin |0.98 |0.95 |0.05| |Prot 2| Na |0.01 |Na |Na |

Note that the endolysin/VAL classifier is one multiclass classifier, implying that their probabilities will always add up to one and that the classifier will always assign one of these to be true.

Also, the endolysin/VAL classifier is only applied to proteins first predicted as being lysins (lysin proba >0.5).

Citation

Boulay, A. et al. PhaLP 2.0: extending the community-oriented phage lysin database with a SUBLYME pipeline for metagenomic discovery. 2025.12.08.692814 Preprint at https://doi.org/10.64898/2025.12.08.692814 (2025).