Selecting Features for Markov Modeling: A Case Study on HP35.

Area of Science:

• Computational biology
• Biophysics
• Protein dynamics

Background:

• Markov state models (MSMs) are crucial for interpreting molecular dynamics simulations.
• Effective MSMs require states that reflect distinct conformations and timescale separation.
• Villin headpiece (HP35) folding is a well-studied model for protein dynamics.

Purpose of the Study:

• To investigate optimal input coordinates ('features') for constructing accurate MSMs.
• To compare the efficacy of dihedral angles versus tertiary contacts for describing protein folding.
• To develop methods for selecting and processing features for improved MSM performance.

Main Methods:

• Analysis of molecular dynamics trajectories for villin headpiece (HP35).
• Comparison of backbone dihedral angles and interresidue distances as input features.
• Development of contact definition and selection strategies for feature engineering.
• Application of low-pass filtering and correlation analysis for state characterization.

Main Results:

• Dihedral angles accurately represent the native state but not the folding process.
• Tertiary contacts provide a superior description of the unfolded states and folding pathway.
• A contact-based MSM accurately reproduces the hierarchical free energy landscape and slow timescales.

Conclusions:

• Feature selection is critical for building mechanistically relevant MSMs.
• Tertiary contacts are essential for capturing the full folding dynamics of proteins like HP35.
• Optimized MSMs using appropriate features enhance the understanding of biomolecular processes.