You can deploy this model and start earning money today!

jpuglia/
ProteinLocationPredictor

joblib

EvolutionaryScale/esmc-300m-2024-12

finetune

mit

Views

No views yet

Protein Location Predictor

Protein Location Predictor

A comprehensive GUI application for predicting protein subcellular localization using SVM and Random Forest classifiers using state-of-the-art protein language models including PROST-T5 and ESM-C embeddings as training data.

Features

Multiple Model Support: Choose from three different prediction models:
- PROST-T5: Transformer-based protein language model
- ESM-C 300M: Evolutionary Scale Modeling (300M parameters)
- ESM-C 600M: Evolutionary Scale Modeling (600M parameters)
User-Friendly GUI: Simple Tkinter-based interface with progress tracking (see screenshot below)
Sequential Processing: Process multiple protein sequences from FASTA files
Flexible Output: Save predictions with confidence scores in text (CSV) format
Error Handling: Comprehensive error handling and user feedback

Supported Python Version

This project has been tested on Python 3.10+.

Requirements

Dependencies (Full environment.yml)

The complete environment definition is located in environment.yml. This file includes all necessary packages for PyTorch, Transformers, ESM models, and GUI operation. Here is a brief excerpt:

yaml

name: tesisEnv
channels:
  - bioconda
  - anaconda
  - conda-forge
  - defaults

# Python version and major packages
dependencies:
  - python=3.10.16
  - pytorch=2.6.0
  - torchvision=0.21.0
  - torchtext=0.18.0
  - transformers=4.46.3
  - scikit-learn=1.6.1
  - biopython=1.85
  - esm=3.1.4
  - numpy=1.26.4
  - joblib=1.4.2
  - tk
  # plus many others (see full file for complete list)

To ensure exact reproducibility, use:

conda env create -f environment.yml

Hardware Requirements

Minimum: 8 GB RAM, CPU-only execution
Recommended: 16 GB+ RAM, NVIDIA GPU with 8 GB+ VRAM
Storage: ~5 GB for model weights and cache

Installation

Clone the repository (with Git LFS for large model files):

bash
1git lfs install 2git clone https://huggingface.co/jpuglia/ProteinLocationPredictor

If you prefer to skip downloading model weights initially:

GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/jpuglia/ProteinLocationPredictor
Navigate into the project directory:

cd ProteinLocationPredictor
Create and activate the Conda environment:

bash
1conda env create -f environment.yml 2conda activate tesisEnv
(If skipped above) Download model weights manually: Model files live in the Models/ directory. If you used GIT_LFS_SKIP_SMUDGE, run:

git lfs pull

Usage

GUI Mode

Launch the application:

python gui.py
In the menu, click File → Load FASTA and select your input file (.fasta, .fa, or .fas).
Choose one of the prediction models (PROST-T5, ESM-C 300M, or ESM-C 600M).
Click Run Prediction and monitor the progress bar.
When complete, you will be prompted to choose an output directory and filename.

Example Input & Output

Input FASTA (example/input.fasta):

>protein_1
MKTVRQERLKSIVRILERSKEPVSGAQLAEELSVSRQVIVQDIAYLRSLGYNIVATPRGYVLAGG
>protein_2
MKTIIALSYIFCLVFAHATAKASEQTDNLQWDLAAIDNSGGHNAVDIKQNLQFQCQNNLHGCF

Output CSV (example/output.csv):

csv

Sequence_ID,Prediction 1,Prediction 2,Prediction 3,Prediction 4,Prediction 5,Prediction 6
protein_1,Cytoplasmic (0.9860),CytoplasmicMembrane (0.0081),Periplasmic (0.0029),Extracellular (0.0019),OuterMembrane (0.0007),Cellwall (0.0003)
protein_2,SignalPeptide (0.7523),Extracellular (0.1234),CytoplasmicMembrane (0.0645),Cellwall (0.0345),Periplasmic (0.0201),OuterMembrane (0.0052)

Project Structure

ProteinLocationPredictor/
├── gui.py
├── src/
│   └── my_utils.py
├── Models/
│   ├── ProstT5_svm.joblib
│   ├── ESMC-300m_svm.joblib
│   ├── ESMC-600m_svm.joblib
│   └── ...
├── environment.yml
├── README.md
└── doc/
    └── screenshots/
        └── gui_example.png

Contributing

Fork the repository
Create a feature branch:

git checkout -b feature/amazing-feature
Commit your changes:

git commit -m "Add amazing feature"
Push to your branch:

git push origin feature/amazing-feature
Open a Pull Request or start a discussion: Repository Discussions

jpuglia/ProteinLocationPredictor

Table of Contents