ProFitFun

ProFitFun evaluates protein model quality by predicting structural fidelity using sequence- and structure-derived features.

Key Features:

• Backbone Dihedral Angle Preferences: Uses backbone dihedral angle preferences at the tripeptide level to capture local conformational propensities.
• Relative Surface Accessibility: Incorporates relative surface accessibility of amino acid residues while accounting for N-terminal and C-terminal neighbors.
• Machine Learning Integration: Applies machine learning algorithms to integrate sequence and structural features for model-quality prediction.

Scientific Applications:

• Validation Datasets: Validated on 25,005 protein structures comprising 23,661 models from 82 non-homologous proteins and 1,344 experimental structures, with an external set of 40,000 models from 200 non-homologous proteins.
• Benchmarking: Benchmarked against four state-of-the-art protein structure quality assessment methods using Spearman's and Pearson's correlation coefficients, average GDT-TS loss, sum of z-scores, and average absolute difference between predicted and observed values.
• Research Applications: Supports computational protein design and research in structural biology, drug discovery, and biotechnology by providing model-quality estimates.

Methodology:

Analysis of sequence- and structure-derived features from experimental protein data, focusing on tripeptide-level backbone dihedral angle preferences and relative surface accessibility, combined with machine learning-based prediction and benchmarking against four state-of-the-art methods using Spearman's and Pearson's correlations, average GDT-TS loss, sum of z-scores, and average absolute difference.