🧬 Molecular Simulation & Protein Structure Prediction

🔬 NMR vs X-ray Crystallography

• NMR (Nuclear Magnetic Resonance)
  • Produces an ensemble of structures (multiple conformations)
  • Reflects protein flexibility in solution
  • Useful for studying dynamic proteins
• X-ray crystallography
  • Produces a single high-resolution structure
  • Requires crystallization (non-physiological)
  • Often misses flexible regions

👉 Key takeaway: NMR = dynamic + multiple conformations X-ray = static + high resolution

⚙️ Protein Folding Problem

• Proteins fold into structures that minimize free energy
• The folding problem = searching for global minimum energy

⚠️ Problem:

• Huge search space (ψ/φ backbone angles + side chains)
• Number of conformations grows exponentially

👉 This is why pure computational folding is hard

🧠 Protein Structure Prediction Methods

1. Ab initio (First principles)

• Uses physics only
• ❌ Too computationally expensive (not practical for large proteins)

2. Comparative Modeling

• Uses known structures as templates

3. Homology Modeling

• Based on sequence similarity
• If sequences are similar → structures are similar

4. Protein Threading

• Fits sequence into known structural folds, even with low similarity

🧵 Protein Threading

💡 Concept

• Place sequence onto known structure
• Evaluate how well it fits (energy scoring)

Why it works:

• Proteins adopt limited number of folds
• ~10 folds explain ~50% of structures

👉 Instead of searching infinite possibilities → reuse known folds

⚙️ Threading Components

• Structural template database
• Energy/scoring function
• Alignment algorithm
• Reliability assessment

📊 Scoring (Punctuation) Functions

Key factors:

• Solvation potentials (buried vs exposed residues)
• Contact potentials
• Secondary structure agreement
• Accessibility predictions

🔥 Contact Potential (Important!)

Uses Boltzmann principle:

• Favorable contacts occur more frequently
• Energy is derived from observed frequencies

👉 Translation:

• Common interactions → energetically favorable

🧬 Sequence Profiles + Secondary Structure

• Combines:
  • Evolutionary info (profiles)
  • Predicted secondary structure

👉 Improves accuracy significantly

📊 Post-processing & Evaluation

• Filter bad models
• Combine additional data
• Benchmark using CASP experiments

🤖 AlphaFold (Deep Learning Revolution)

🧠 Overview

• Uses deep learning + evolutionary data
• Predicts 3D structure from sequence

🧬 Key Components

1. Input

• Amino acid sequence

2. MSA (Multiple Sequence Alignment)

• Detects:
  • conserved residues
  • co-evolution (residues interacting)

👉 If two residues mutate together → likely interact

3. Evoformer (Core engine)

• Transformer-based
• Learns:
  • long-range interactions
  • structural constraints

4. Structure Module

• Converts predictions into 3D coordinates
• Uses Invariant Point Attention (IPA)

👉 Important:

• Handles spatial geometry properly

5. Output

• Full atomic model
• Confidence metrics:
  • pLDDT → per residue confidence
  • PAE → domain relationship uncertainty

🧪 Molecular Docking

🔑 What is Docking?

• Predicts how a ligand binds to a protein

⚠️ Important:

• Does NOT directly predict bioactivity

🔄 Docking Theories

• Lock-and-key → rigid fit
• Induced fit → protein adapts

⚙️ Two Main Steps

1. Sampling

• Try many ligand conformations

2. Scoring

• Rank based on binding energy

🧬 Types of Docking

• Protein–protein
• Protein–ligand
• Protein–nucleotide

🎯 Applications

• Reproduce known binding modes
• Predict binding of known ligands
• Estimate binding affinities
• Virtual screening (drug discovery)

⚙️ Algorithms

• Use:
  • conformational search methods
  • scoring functions

🔍 Docking Workflow

Typical steps:

1. Prepare protein + ligand
2. Define binding site
3. Generate conformations
4. Score poses
5. Rank results

🧪 Validation Methods

🔁 Redocking

• Dock ligand back into known structure
• Tests accuracy

🔄 Cross-docking

• Dock ligand into different structures
• Tests robustness

📈 ROC Curve (Image slide explanation)

• Measures model performance
• AUC (Area Under Curve):
  • 1.0 = perfect
  • 0.5 = random

👉 Used in virtual screening

🔬 Binding Energy Example

• Example: -10.10 kcal/mol
• More negative = stronger binding

🧾 Docking Score Table

• Compares different ligand poses
• Used to select best candidate

⚠️ Types of Docking (Advanced)

• Covalent docking → ligand forms bond
• Blind docking → unknown binding site
• Reverse docking → one ligand vs many proteins

⚖️ Pros & Cons of Docking

✅ Pros

• Fast screening
• Cost-effective
• Useful for drug discovery

❌ Cons

• Scoring functions imperfect
• Protein flexibility limited
• Does not guarantee biological activity

🧠 Big Picture Summary

• Protein structure prediction evolved from:
  • ❌ physics-based → too slow
  • ✅ template-based → useful
  • 🚀 AI-based (AlphaFold) → breakthrough
• Docking:
  • Predicts binding mode, not activity
  • Depends heavily on sampling + scoring quality

🔥 Key Concept Connections (Important for Exams)

• Threading vs Homology modeling
  • Homology → sequence similarity
  • Threading → structure similarity
• AlphaFold vs Docking
  • AlphaFold → structure prediction
  • Docking → interaction prediction
• Energy principle
  • Folding → minimize energy
  • Docking → optimize binding energy