Searching for Structure: Characterizing the Protein Conformational Landscape with Clustering-Based Algorithms

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• Protein conformation identification is a complex, high-dimensional challenge.
• Understanding protein dynamics is crucial for biological function.
• Existing clustering methods have limitations in characterizing complex protein landscapes.

Purpose of the Study:

• To evaluate the Secondary Structural Ensembles with machine Learning (StELa) double-clustering method.
• To compare StELa's performance against RMSD-based and CATS clustering algorithms.
• To assess the ability of StELa to identify key features of protein free energy landscapes (FELs).

Main Methods:

• StELa clusters protein structures using dihedral angles (φ and ψ) and secondary structure.
• States are classified as vectors of cluster indices from Ramachandran plots.
• Hierarchical clustering of vectors identifies FEL features.
• Comparison with RMSD-based (global properties) and CATS (dihedral angle distributions) methods.

Main Results:

• StELa successfully identifies minima and relevant energy states on FELs for various protein types.
• RMSD-based clustering produced too many clusters, obscuring distinct states.
• CATS struggled to adequately sample FELs for long intrinsically disordered proteins (IDPs) and globular protein fragments.

Conclusions:

• StELa is a superior method for characterizing protein conformational ensembles and their free energy landscapes.
• StELa accurately captures local structural properties relevant to protein dynamics.
• The method shows promise for analyzing complex systems like IDPs and protein fragments.