HLA-B39 Simulation Analysis

Analysis pipeline for molecular dynamics simulation data of HLA-B39 subtypes and B38:01 in peptide-loaded and peptide-free states.

Overview

This repository analyzes protein interaction energies and network centrality changes in HLA class I molecules using molecular dynamics simulation data. The analysis focuses on identifying residues and interaction pairs that respond to peptide binding and differ between allelic variants.

Repository Structure

Analysis Scripts

1_preprocessing.py

Preprocessing pipeline that processes molecular dynamics simulation data from gRINN outputs.

Steps:

1. Load interaction energy data from gRINN outputs
2. Annotate residue pairs with structural information
3. Identify interaction pairs with consistent energies across replicates (ANOVA)
4. Compare peptide-loaded vs peptide-free systems (t-test)
5. Compare B39 subtypes with B38:01 reference
6. Construct protein energy networks (Ribeiro-Ortiz method)
7. Calculate betweenness centrality for all residues
8. Identify residues with significant centrality changes upon peptide binding

Output: 8 CSV files containing preprocessed datasets

Runtime: Several minutes (network construction is computationally intensive)

2_analysis.ipynb

Jupyter notebook for visualization and exploratory analysis.

Sections:

• Summary statistics of interaction changes by allele and peptide
• 3D structure visualization of significantly affected residue pairs
• Conservation score integration with network centrality analysis
• Polymorphic position analysis across alleles

Prerequisites: Run 1_preprocessing.py first to generate required CSV files

Workflow

# Step 1: Run preprocessing (once)
python3 1_preprocessing.py

# Step 2: Run analysis (interactive)
jupyter notebook 2_analysis.ipynb

Data Requirements

The preprocessing script expects gRINN outputs at:

BASE_FOLDER/*/grinn_output_skip10/energies_intEnVdW.csv
BASE_FOLDER/*/grinn_output_skip10/system_dry.pdb

Methods

Statistical Tests:

• ANOVA: Identify consistent interactions/residues across replicates (p > 0.05)
• t-test: Compare peptide-loaded vs peptide-free conditions (p < 0.05)
• Energy threshold: Absolute mean difference > 2 kJ/mol

Network Analysis:

• Protein energy networks constructed using Ribeiro-Ortiz method
• Betweenness centrality calculated for all residues across simulation frames
• Equilibration: Frames > 20 retained for analysis

Output Files

File	Description
intEnVdW_2025_07_10.csv	Raw interaction energies with annotations
df_cons_pairs_saved_2025_07_10.csv	Consistent interaction pairs (ANOVA p > 0.05)
df_sig_aff_pairs_loaded_2025_07_10.csv	Pairs affected by peptide loading
df_pairs_only_in_pmhc_2025_07_10.csv	Pairs only present in peptide-loaded systems
df_sig_aff_pairs_b39_2025_07_10.csv	Pairs differing between B39 and B38:01
df_bc_equil_2025_07_10.csv	Betweenness centrality values (equilibrated frames)
df_bc_equil_cons_resids_2025_07_10.csv	Residues with consistent BC across replicates
df_bc_sigaff_resids_2026_01_27.csv	Residues with significant BC changes

Dependencies

prody
pandas
numpy
scipy
matplotlib
seaborn
networkx
py3Dmol
natsort
tqdm