An MML classification of protein structure that knows about angles and sequence

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Bioinformatics

Background:

• Protein secondary structure classification is crucial for understanding protein function.
• Previous methods like AutoClass and Snob have been used for protein data clustering.
• Representing protein residue data using dihedral angles (phi and psi) is more efficient than Cartesian coordinates.

Purpose of the Study:

• To develop a more appropriate model for protein secondary structure classification using Hidden Markov Models (HMMs).
• To improve upon existing mixture and Markov modeling techniques for serial correlation in protein data.
• To identify and characterize distinct protein structure classes and their relationships.

Main Methods:

• Utilized the MML classification program, Snob, extended for Markov modeling of circular data.
• Applied a Hidden Markov Model to cluster protein dihedral angle data (phi and psi).
• Analyzed the class transition matrix and related the identified classes back to amino acid sequences.

Main Results:

• The Hidden Markov Model resulted in a simpler classification model with 17 distinct structure classes.
• Identified 3 types of Helix, 2 types of Beta structures, and multiple types of Turn structures.
• The most numerous Turn class represents a continuous flexible structure negatively correlated with other classes.

Conclusions:

• Hidden Markov Models provide a more appropriate framework for modeling protein secondary structure data.
• The developed model offers a refined classification of protein structures, revealing nuanced relationships between different secondary structure types.
• The findings highlight the utility of HMMs in uncovering complex patterns in biological sequence and structure data.