Disulfide Replace: Generative Design of Peptides using Rosetta

Elevator Pitch: This repository contains the computational pipeline for the Anti-HIV scFv project, focusing on the generative design of peptides using Rosetta. It automates the transition from sequence generation to high-resolution energy minimization and statistical stability ranking, enabling the replacement of disulfide bonds with optimized hydrophobic packing.

Quick Start

Since Rosetta environments are complex, we provide streamlined entry points. Assuming Rosetta is installed and $ROSETTA3 is set:

# 1. Identify disulfide neighbors (Group 1 & 2)
pymol -qc scripts/pymol/search_within_4A.pml

# 2. Generate combinatorial hydrophobic mutants
pymol -qc scripts/pymol/generate_mutants.pml

# 3. High-resolution scoring and relaxation (parallel)
./scripts/rosetta/runit_fastrelax_parallel.sh

# 4. Perform Bootstrap Statistical Analysis
python scripts/analysis/bootstrap_fastrelax_analysis.py

# 5. Visualize results
python scripts/analysis/plot_fastrelax_results.py

Reproducibility Note

To ensure 100% reproducibility of the results presented in our research:

• Energy Minimization: Performed with a constant seed (-run:jran 11105) to prevent stochastic variance during initial optimization.
• FastRelax: Performed using the standard ref2015 score function across all steps. Replicates use independent random seeds to ensure diverse sampling of the local energy landscape.
• Bootstrap Stability: Replicates are ranked based on their converged mean energy after ensemble relaxation.

Repository Organization

The project is structured to separate inputs, logic, and results for a professional scientific workflow:

• inputs/: Source protein structures (e.g., 5B3N.cif) and generated mutants.
• scripts/ (Pipeline Logic):
  • pymol/: Identification and mutagenesis scripts (.pml).
  • rosetta/: Production Rosetta execution scripts (.sh).
  • analysis/: Plotting and statistics scripts (.py).
• results/:
  • figures/: Professional plots and visualizations.
  • tables/: Summary CSV files with aggregated rankings and bootstrap statistics.
  • rosetta_fastrelax_results/: Relaxation output and diagnostic logs (capturing seeds and scores).
  • rosetta_minimizeenergy_results/: Minimization output and diagnostic logs.
• metadata:
  • DATA.md: Detailed data dictionary for CSV columns and REU units.
  • requirements.txt: Python environment specifications.

Scientific Workflow

1. Neighbor Identification: Identify disulfide neighbor Leu, Ile, Met residues within 4Å of the disulfide sidechains.
2. Combinatorial Search: Construct a search considering Ala/Val at Cys sites and Leu/Ile/Met at neighboring sites.
3. Hydrophobic Optimization: Combinations are filtered to maintain or reduce total heavy atom count relative to wildtype.
4. Ensemble Evaluation: Mutants are ranked by the mean score across an ensemble of 20 FastRelax replicates.
5. Bootstrap Convergence: We use m-out-of-n bootstrap analysis ($N=20$) to estimate the precision of our relaxation protocol. By sampling 20 replicates over 1000 iterations, we confirm that our standard 20-rep protocol is sufficient for the energy mean to converge, ensuring robust identification of stable mutants.

Software Citations

Rosetta

"Computational modeling and design were performed using Rosetta version 2023.45 (release a6d9ba8). Leman, J. K., et al. (2020)."

PyMOL

"Molecular visualizations were generated using The PyMOL Molecular Graphics System, Version 3.1.3, Schrödinger, LLC."