MEDUSA

MEDUSA predicts residue-level protein flexibility from amino acid sequences using a convolutional neural network that integrates evolutionary and physico-chemical information to assign normalized B-factor-based flexibility classes.

Key Features:

• Deep Learning Framework: Employs a convolutional neural network (CNN) that analyzes input derived from amino acid sequences while leveraging evolutionary information from homologous protein sequences and physico-chemical properties of amino acids.
• Flexibility Prediction Classes: Provides multiclass flexibility predictions in two, three, and five class schemes based on expected normalized B-factor values.
• Training Dataset: Trained on a comprehensive, non-redundant dataset of X-ray crystallography-derived protein structures.

Scientific Applications:

• Understanding Molecular Mechanisms: Predicts residue flexibility to inform analyses of protein stability, intermolecular interactions, and functional dynamics.
• Identification of Dynamic Regions: Identifies potentially highly deformable regions within protein sequences to characterize local dynamic properties.
• Complementary Structural Biology Analyses: Supplies flexibility estimates where experimental structural data are limited, supporting interpretation of protein behavior.

Methodology:

Integrates evolutionary information from homologous protein sequences and physico-chemical amino acid properties into a convolutional neural network trained on a non-redundant set of X-ray crystallography-derived protein structures to assign per-residue flexibility classes based on normalized B-factor values.