MASS-p

MASS-p assesses the global quality of individual protein structural models using six novel statistical potentials combined with a random forest ensemble (Model Assessment by Statistical Scores).

Key Features:

• Single-Model Input: Operates on individual protein models without requiring a pool of decoys.
• Random Forests and Novel Statistical Potentials: Uses a random forest ensemble with six novel energy functions (statistical potentials) as primary predictive features.
• Comparative Feature Importance: The novel potentials show higher importance than existing energy functions such as RWplus, GOAP, DFIRE, and Rosetta when used as machine learning features.
• Benchmark Evaluation Metrics: Evaluated against top-performing single-model methods from CASP11 and reported comparable performance in CASP12 and CASP13 across the four official evaluation criteria: assigning relative and absolute scores, selecting the best model from decoys, and discriminating good versus bad models.

Scientific Applications:

• Model selection from decoys: Identifies the best structural models within decoy sets when only single models are available.
• Protein structure prediction assessment: Provides global quality scores to support protein structure prediction, interpretation of protein function, and analysis of protein interactions.

Methodology:

MASS-p computes six novel statistical energy functions and integrates them as features into a random forest machine learning model to predict global model quality.