Folding Free Energy (ΔΔG)

ΔΔG is defined as ΔG_folding(mutant) - ΔG_folding(wild-type). A negative ΔΔG indicates the mutation stabilizes the protein (more favorable folding), while a positive ΔΔG indicates destabilization. The measurement reflects changes in enthalpy (hydrogen bonding, van der Waals) and entropy (conformational freedom, hydrophobic effect) upon mutation.

Mutations alter stability through several mechanisms:

• Cavity creation: Removing a buried hydrophobic sidechain creates energetically unfavorable empty space
• Steric clashes: Introducing a larger residue in a packed core causes atomic overlap
• Electrostatic disruption: Altering charge networks or salt bridges at the protein surface
• Backbone strain: Proline or glycine substitutions that perturb local phi/psi angle preferences

Modern ΔΔG predictors fall into several categories:

• Physics-based force fields: FoldX, Rosetta ddg_monomer — use empirical energy functions with dielectric models and side-chain repacking
• Statistical potentials: PoPMuSiC, SDM — derive residue pair preferences from known structures
• Machine learning: ThermoNet, DDGun — train on experimental databases like ProTherm
• Protein language models: ESM-1v, Tranception — leverage evolutionary sequence context without explicit structural input

Predicted ΔΔG values are benchmarked against:

• Differential scanning calorimetry (DSC): Directly measures heat capacity changes during thermal denaturation
• Chemical denaturation: Monitors unfolding via tryptophan fluorescence or circular dichroism with urea or guanidinium chloride
• Deep mutational scanning: High-throughput fitness assays that provide stability proxies at scale
• ProTherm database: The curated reference set of experimentally measured ΔΔG values used for training and validation

Prediction accuracy is assessed using:

• Pearson correlation coefficient (r): Measures linear agreement between predicted and experimental ΔΔG; state-of-the-art methods achieve r ≈ 0.5–0.7 on blind benchmarks
• RMSE (Root Mean Square Error): Typical values range from 1.0–1.5 kcal/mol
• Classification accuracy: Binary discrimination of stabilizing vs. destabilizing mutations, often exceeding 80%
• Antisymmetry: A critical test checking that ΔΔG(A→B) ≈ -ΔΔG(B→A); many methods fail this consistency check

ΔΔG prediction drives rational design strategies:

• Thermostabilization: Identifying mutations that rigidify flexible loops or optimize core packing for industrial enzyme applications
• Affinity maturation: Predicting mutations that stabilize the bound conformation of an antibody without affecting the unbound state
• Variant effect interpretation: Classifying missense mutations as benign or pathogenic based on predicted destabilization magnitude
• Solubility engineering: Reducing aggregation propensity by optimizing surface charge distribution

The thermodynamic potential measuring the difference in free energy between a protein's folded native state and its unfolded ensemble. A negative ΔG indicates spontaneous folding under given conditions.

• ΔG = G_folded - G_unfolded: Typically ranges from -5 to -15 kcal/mol for stable globular proteins
• Components: Balances favorable enthalpy (hydrogen bonds, van der Waals) against unfavorable entropy loss
• Measurement: Experimentally determined via chemical denaturation monitored by circular dichroism or fluorescence
• Context: ΔΔG is always calculated relative to this baseline wild-type ΔG value

Community benchmarks that systematically evaluate ΔΔG prediction methods against blinded experimental data to track progress and identify methodological gaps.

• Key Metrics: Pearson and Spearman correlation coefficients, root mean square error (RMSE), and classification accuracy for stabilizing vs. destabilizing
• Anti-Symmetric Bias: Many predictors systematically underestimate destabilization magnitude for reverse mutations (A→B vs. B→A)
• Current Limits: Top methods achieve correlations of ~0.6-0.7; predicting stabilizing mutations remains harder than destabilizing ones
• Datasets: S2648, S669, and p53 benchmarks are standard evaluation sets curated from ProTherm and FireProtDB