ProMaya

ProMaya predicts the impact of single-site amino-acid substitutions on protein thermal stability by estimating the folding free-energy change (ΔΔG) between mutant and wild-type proteins.

Key Features:

• Collaborative-filtering baseline model: Leverages experimental ΔΔG measurements from known mutations at the same position and other positions to inform predictions.
• Random-forest regression: Uses an ensemble of decision trees to model non-linear relationships among features and improve robustness.
• Diverse feature set: Incorporates a broad set of predictive features, including outputs from two established stability-prediction methods, to improve generalization across proteins and mutation contexts.
• Accurate predictions: Reported validation performance includes Pearson correlation of 0.79 and RMSE of 0.96.
• Effective use of prior data: Exploits existing position-specific ΔΔG measurements to improve reliability of predictions for other substitutions at the same site.
• Versatility across data regimes: Maintains competitive performance when position-specific experimental data are unavailable, supported by its comprehensive feature representation.

Scientific Applications:

• Disease-variant interpretation: Interpreting the molecular basis of disease-associated variants linked to protein destabilization.
• Protein design and engineering: Guiding protein design and engineering for improved stability in therapeutic and industrial settings.
• Experimental prioritization: Prioritizing mutations for experimental testing by providing computational estimates of stability effects.

Methodology:

Predicts the stability free-energy difference (ΔΔG) between mutant and wild-type proteins by integrating a collaborative-filtering baseline that leverages experimental ΔΔG measurements at the same and other positions, random-forest regression (an ensemble of decision trees), and a diverse feature set including outputs from two established stability-prediction methods.