Module 5

Protein Structure & AlphaFold2

Protein structure prediction was the “second half-century problem” of biology until AlphaFold2 essentially solved it in 2020. This module reviews the pre-AF2 methods (homology, threading, ab initio), the AlphaFold2 architecture (Evoformer + Structure Module), and the AlphaFold-DB era of practical use.

1. Pre-AF2 Methods

Homology modelling (Modeller, SWISS-MODEL) copies coordinates from a related experimentally-solved structure. Threading (I-TASSER, RaptorX) aligns to fold libraries even without close homology. Ab initio (Rosetta, QUARK) samples conformations against a force-field + statistical potential. Before 2020, CASP competitions showed slow but steady progress; accuracy on novel folds plateaued around GDT-TS 40–60.

2. AlphaFold2 Architecture

Jumper 2021 (Nature) described AF2. Core components:

MSA + template search: recover co-evolutionary and structural context from UniProt / BFD / PDB.
Evoformer: 48 blocks of attention that iteratively refine MSA embeddings and pair representations.
Invariant Point Attention (IPA) Structure Module: 8 blocks that produce 3-D coordinates + atomic-level refinement.
Recycling: the whole network is run 3× with iterated inputs for refinement.

Simulation: Accuracy & Confidence

Python

script.py42 lines

import numpy as np, matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

# CASP accuracy history (GDT-TS, higher = better)
years = np.arange(2006, 2023, 2)
casp_median = [46, 48, 50, 52, 55, 58, 60, 65, 92]
participants = 'top method'

# AlphaFold2 pLDDT confidence distribution
rng = np.random.default_rng(0)
plddt_good = rng.beta(8, 2, 300) * 100
plddt_bad  = rng.beta(3, 4, 100) * 100
plddt = np.concatenate([plddt_good, plddt_bad])

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(14, 5), facecolor='#0a0a1a')
for ax in (ax1, ax2):
    ax.set_facecolor('#111827'); ax.tick_params(colors='#cbd5e1')
    for s in ax.spines.values(): s.set_color('#334155')
    ax.grid(True, color='#334155', alpha=0.3)

ax1.plot(years, casp_median, color='#2dd4bf', lw=2.6, marker='o', markersize=8)
ax1.axvline(2020, color='#fbbf24', ls='--', label='AlphaFold2 (CASP14)')
ax1.set_xlabel('CASP year', color='#cbd5e1')
ax1.set_ylabel('Best-method median GDT-TS', color='#cbd5e1')
ax1.set_title('Protein-structure prediction accuracy',
              color='#5eead4', fontweight='bold')
ax1.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')

ax2.hist(plddt, bins=30, color='#2dd4bf', edgecolor='#5eead4')
ax2.axvline(90, color='#34d399', ls='--', label='Very high (>90)')
ax2.axvline(70, color='#fbbf24', ls='--', label='Confident (>70)')
ax2.set_xlabel('pLDDT', color='#cbd5e1')
ax2.set_ylabel('Residues', color='#cbd5e1')
ax2.set_title('Per-residue pLDDT confidence',
              color='#5eead4', fontweight='bold')
ax2.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')

plt.tight_layout()
plt.savefig('output.png', dpi=120, bbox_inches='tight', facecolor='#0a0a1a')
print('CASP14 2020: AlphaFold2 median GDT-TS 92.4 vs runner-up 69.4')
print('pLDDT > 90 = experimental-quality; 70-90 confident; <50 low confidence')

Click Run to execute the Python code

Code will be executed with Python 3 on the server

3. AlphaFold-DB & Practical Use

AlphaFold Database (Varadi 2022) hosts predicted structures for ~200 million proteins — essentially all of UniProt. pLDDT (per-residue confidence), PAE (predicted aligned error) between residue pairs, and global pTM are used for quality control. Downstream uses: structure-based drug discovery, docking (HADDOCK, AutoDock Vina), variant-effect prediction, metagenome annotation. AlphaFold3 (Abramson 2024) extended to protein-protein, protein-DNA, protein-ligand with unified attention over multi-modal inputs.

Key References

• Jumper, J. et al. (2021). “Highly accurate protein structure prediction with AlphaFold.” Nature, 596, 583–589.

• Varadi, M. et al. (2022). “AlphaFold Protein Structure Database.” Nucleic Acids Res., 50, D439–D444.

• Abramson, J. et al. (2024). “Accurate structure prediction of biomolecular interactions with AlphaFold 3.” Nature, 630, 493–500.

• Lin, Z. et al. (2023). “Evolutionary-scale prediction of atomic-level protein structure.” Science, 379, 1123–1130.

Share:X Reddit LinkedIn

← Module 4 Module 6: Phylogenetics →