Epitope Immunogenicity Classifier

A semi-supervised Gaussian Mixture Model for predicting peptide immunogenicity using Kidera factors.

Overview

This project replicates and extends an R-based immunogenicity classifier using Python. It uses:

• Kidera factors (10 physicochemical properties) for amino acid encoding
• Gaussian Mixture Model with semi-supervised initialization
• Two feature calculation modes: direct embedding or sliding window summation

Installation

git clone https://github.com/AnatoliyLarkin/pyipred.git
pip install -r requirements.txt

Quick Start

from model_semisupervised import GMM

# Initialize and load features
mod = GMM()
mod.load_features('datasets/kidera.txt')

# Load training data
mod.load_training_data(X, Y, 'full')  # or 'sliding_window' with k parameter
mod.train()

# Predict
prob, class_ = mod.predict('ELALGIGILV', 'full')

Please refer to semisupervised_demonstration.ipynb for detailed tutorial

Key Components

• model_semisupervised.GMM: Main classifier class

• calculate_features.py: Amino acid feature processing

• Semi-supervised EM initialization using labeled data

• Automatic threshold optimization (Youden's index)

References

1. Pogorelyy, M. V. et al. Exploring the pre-immune landscape of antigen-specific T cells. Genome Med 10, (2018).
2. Buckley, P. R. et al. Evaluating performance of existing computational models in predicting CD8+ T cell pathogenic epitopes and cancer neoantigens. Briefings in Bioinformatics 23, (2022).
3. Kidera, A., Konishi, Y., Oka, M., Ooi, T. & Scheraga, H. A. Statistical analysis of the physical properties of the 20 naturally occurring amino acids. J Protein Chem 4, 23–55 (1985).