Menu

Oxford TODO App

MOSAICS WP1 Tutorial

Methodologies for Optimization and SAmpling In Computational Studies

Master advanced Monte Carlo sampling methods for molecular simulation. Become an expert in Parallel Tempering and Equi-Energy Monte Carlo.

Progress: 0/6 modules complete

📖 Case Study Reference

📌 Keep this PDF open as reference throughout the tutorial. It contains essential background and experimental details.

The MOSAICS Framework

MOSAICS = Methodologies for Optimization and SAmpling In Computational Studies.Created by Prof. Peter Minary at Stanford (2007), continuously developed at Oxford. A revolutionary software framework for modeling biomolecular structures using advanced Monte Carlo sampling and optimization.

Key Innovation

Uses natural degrees of freedom instead of Cartesian coordinates

Speed Advantage

Minutes vs days compared to traditional MD simulations

🧬 Critical Need for CRISPR Research

The Data Gap: Training GNN models for CRISPR-Cas9 off-target prediction requires 3D structures of Cas9 complexes with different DNA-RNA mismatches. Current available data: ~30 structures. Needed: thousands.

Traditional MD Limitations:

  • Extremely long timescales (days to weeks per structure)
  • Gets trapped in local energy minima
  • Cannot efficiently sample conformational space
  • Prohibitively expensive for high-throughput structure generation

MOSAICS Solution: Generate relaxed, biologically relevant Cas9-DNA-RNA structures in MINUTES, not days, enabling rapid dataset creation for ML training.

🎯 Peter's Direction 3: The Pipeline

📝
Any mismatch pair
MOSAICS relaxation
🔬
Extract features
🧠
GNN prediction

🚧 The Two Fundamental Problems

1. DIMENSIONALITY PROBLEM

Issue: Too many degrees of freedom (3N for N atoms)

Solution: WP2-4 address this with natural coordinates and constraints

Uses bond angles, dihedrals instead of x,y,z coordinates

2. ENERGY SURFACE PROBLEM

Issue: Deep local minima trap simulations

Solution: WP1 - Advanced sampling (PT, EEMC)

Temperature exchanges help escape energy traps

WP1 (this tutorial) solves the energy surface problem

📚 Essential References

Foundational Papers:

  • • Minary (2007) MOSAICS Software Framework
  • • Sim, Levitt, Minary (2012) PNAS 109(8):2890-2895

Core Methods:

  • • Geyer (1991) Parallel Tempering
  • • Metropolis et al. (1953) Monte Carlo Method