biofold/is-pero repository overview

⁠Is-Pero: Peroxisomal Protein Localization Prediction

A comprehensive web server for end-to-end prediction of peroxisomal protein localization from raw FASTA sequences or UniProt identifiers.

Web Server: https://structure.biofold.org/is-pero/⁠

⁠Overview

Is-Pero integrates four machine learning modules to predict peroxisomal protein localization and characterization:

1. Binary Classification - Distinguishes peroxisomal vs. non-peroxisomal proteins
2. Sub-compartment Prediction - Assigns matrix or membrane localization
3. Signal Detection - Identifies PTS1, PTS2, and membrane-targeting signals
4. PEX19 Binding Prediction - Predicts membrane-targeting sites

All models use ESM-2 protein language model embeddings and handle extreme class imbalance through SMOTE balancing.

⁠Installation

⁠Docker (Recommended)

docker pull biofold/is-pero
mkdir -p ./output
docker run -it \
    --user $(id -u):$(id -g) \
    -v "$(pwd)/output:/workspace/is-pero/output" \
    -e HOME=/workspace/is-pero/output \
    biofold/is-pero \
    python /workspace/is-pero/pero-cascade.py \
        --fasta /workspace/is-pero/test/web_sequences.fasta \
        --output /workspace/is-pero/output/web \
        --filter

⁠Local Setup

git clone https://github.com/biofold/is-pero.git
cd is-pero
conda env create -f is-pero-env.yml -n is-pero

⁠Usage

⁠Web Interface

1. Go to https://structure.biofold.org/is-pero/⁠
2. Submit FASTA file or UniProt ID
3. Download CSV results

⁠Command Line

conda activate is-pero
python pero-cascade.py --fasta test/web_sequences.fasta --output output/web --filter

⁠Input Format

FASTA file:

>sp|P0C024|NUDT7_HUMAN
MGLSDGEWQLVLNVWGKVEADIPGHGQEVLIRLFKGHPETLEK...

>sp|P21549|AGT1_HUMAN
MAGVLGLGPWRLFPNADHPFLFFGRGAGGAAAAGALALGKA...

UniProt IDs:

P0C024,P21549,P28288

⁠Output

Results are provided as CSV with the following columns:

• Sequence_ID - Protein identifier

• Peroxisomal - Classification (yes/no)

• Probability_Peroxisomal - Confidence score (0-1)

• FDR-pero - Conservative FDR (Probability_Peroxisomal ≥ 0.5) or FOR (Probability_Peroxisomal < 0.5).

• Matrix or Membrane - Sub-compartment assignment

• Probability_Matrix - Sub-compartment confidence (0-1)

• FDR-matrix - Conservative FDR (Probability_Matrix ≥ 0.5) or FOR (Probability_Matrix < 0.5)

• Peptide - Detected targeting signal

• Start - Signal position in protein

• Score - PSSM score for signal

• Signal - Signal type (PTS1/PTS2/mPTS)

• Probability_mPTS - mPTS binding probability (0-1)

• FDR-mpts Conservative FDR (Probability_MPTS ≥ 0.5) or FOR (Probability_MPTS < 0.5)

Example output:

Sequence_ID,Peroxisomal,Probability_Peroxisomal,FDR-pero,Matrix,Probability_Matrix,FDR-matrix,Peptide,Start,Score,Signal,Probability_mPTS,FDR-mpts
B3DMA2,1,9.5e-01,3.7e-01,matrix,9.7e-01,2.5e-01,ARM,777,N/A,PTS1,N/A,N/A
B8AME2,1,9.8e-01,2.2e-01,matrix,9.5e-01,2.5e-01,PSM,490,N/A,PTS1,N/A,N/A
Tb927.10.1860:mRNA,1,5.1e-01,6.8e-01,membrane,1.0e-01,1.7e-01,LGRLRRYVNDLLATN,297,7.4e-01,not-mPTS,3.4e-01,1.8e-01
Tb927.10.1860:mRNA,1,5.1e-01,6.8e-01,membrane,1.0e-01,1.7e-01,ACVIIAIILLCILAR,320,8.4e-01,mPTS,1.0e+00,0.0e+00
Tb927.10.8410:mRNA,1,7.4e-01,5.6e-01,membrane,3.5e-03,1.7e-01,LSLTCLLLDEVLLLR,136,6.9e-01,mPTS,1.0e+00,0.0e+00
Tb927.2.4130:mRNA,1,8.2e-01,5.0e-01,matrix,5.0e-01,2.5e-01,N/A,N/A,N/A,N/A,N/A,N/A
Tb927.3.5090:mRNA,0,1.4e-01,4.1e-03,N/A,N/A,N/A,N/A,N/A,N/A,N/A,N/A,N/A
Tb927.9.1720:mRNA,1,9.6e-01,3.4e-01,membrane,3.5e-04,0.0e+00,YKILFDALKSFRNGF,32,7.2e-01,mPTS,8.8e-01,0.0e+00

⁠Performance

⁠Requirements

• Python 3.8+
• PyTorch 1.12+
• scikit-learn 1.0+

See requirements.txt for full dependencies.

⁠Citation

@article{ispero2025,
  title={Is-Pero: a webserver application for predicting the localization of peroxisomal proteins},
  author={Anteghini, Marco and Krishna, Chethan K. and Kalel, Vishal C. and Zauli, Andrea and dos Santos, Vitor M. and Saccenti, Edoardo and Erdmann, Ralf and Capriotti, Emidio},
  journal={},
  year={2025},
  url={https://structure.biofold.org/is-pero/}
}

⁠License

Scripts are licensed under the Creative Commons by NC-SA 4.0 license.

⁠Support

• Email: [email protected]⁠
• Web Server: https://structure.biofold.org/is-pero/⁠
• GitHub: https://github.com/biofold/is-pero⁠

⁠Corresponding Authors

• Marco Anteghini (BioFolD, University of Bologna)
• Emidio Capriotti (BioFolD, University of Bologna)