Foldtree_ProstT5

⚠️ CAUTION: This pipeline has not been benchmarked yet. Use at your own risk and validate results carefully.

A Snakemake pipeline that provides a Foldtree replacement for phylogenetic tree construction when protein structures are not available. This pipeline leverages ProstT5 embeddings through Foldseek to generate statistically corrected and rooted sequence identity trees.

Overview

Foldtree_ProstT5 is designed for scenarios where:

• Protein structures are unavailable for your sequences of interest
• You need phylogenetic trees based on structural similarity estimates
• Traditional Foldtree cannot be used due to lack of structural data

The pipeline operates in "fident mode" (sequence identity mode) and provides:

• ✅ Statistically corrected sequence identity trees
• ✅ Rooted phylogenetic trees
• ❌ Does NOT output LDDT distance matrices
• ❌ Does NOT output TM-score distance matrices

Prerequisites

Required Software

• Snakemake (≥7.0)
• Foldseek
• Conda/Mamba for environment management

Required Data

You must download the ProstT5 weights before running this pipeline:

# Download ProstT5 weights using Foldseek
foldseek databases ProstT5 weights tmp

Important: Ensure you have sufficient disk space (~XX GB) for the ProstT5 database.

Quick Start

1. Installation

# Clone the repository
git clone https://github.com/DessimozLab/Foldtree_ProstT5.git
cd Foldtree_ProstT5

# Install Snakemake (if not already installed)
conda install -c bioconda snakemake

# Set up the project structure
make setup

2. Download ProstT5 Weights

# This may take several hours and requires significant disk space
foldseek databases ProstT5 data/prostT5_weights tmp

make sure to set the correct path in the config file at workflow/config/config_vars.yaml:

prostt5_weights: /path/to/your/Foldtree_ProstT5/data/prostT5_weights

3. Prepare Input Data

Place your protein sequences in FASTA format in data/sequences/:

cp your_sequences.fasta data/sequences/

5. Run the Pipeline

#run with Snakemake
snakemake -s workflow/rules/fold_tree_prostT5 --use-conda --cores 4 --config folder=./data/sequences 

Pipeline Workflow

1. Sequence Preprocessing: Validates and filters input sequences
2. ProstT5 Embedding: Generates structural embeddings using Foldseek + ProstT5
3. Distance Calculation: Computes sequence identity from embeddings
4. Statistical Correction: Applies evolutionary distance corrections
5. Tree Construction: Builds initial phylogenetic tree from distance matrix
6. Tree Rooting: Roots the tree using specified method

Limitations & Important Notes

⚠️ Critical Limitations:

• No LDDT matrices: This pipeline cannot output LDDT-based distance matrices
• No TM-score matrices: TM-score calculations are not supported
• Fident mode only: Only operates in sequence identity mode, not structural similarity mode
• Requires ProstT5 weights: Must download large database files (~XX GB)
• Not benchmarked: Results should be validated against known phylogenies when possible

Cluster Execution

# For SLURM clusters
snakemake --cluster-config config/cluster_config.yaml \
          --cluster "sbatch --partition=normal --time=4:00:00" \
          --jobs 20 --use-conda --directory #your dataset path#

Troubleshooting

Common Issues

1. ProstT5 database missing: Ensure you've downloaded the weights using foldseek databases
2. Memory errors: Increase memory allocation in cluster config for large datasets
3. Slow performance: Use more cores or consider splitting large sequence sets

Citation

If you use this pipeline, please cite:

• Foldseek: [Steinegger & Söding, 2022]
• ProstT5: [Heinzinger et al., 2023]
• Foldtree : [Moi et al., 2025]

Contributing

This pipeline is under active development. Please report issues or contribute improvements via GitHub.

License

This project is licensed under the MIT License. See the LICENSE file for details.