AMUSET-TICA: A Tensor-Based Approach for Identifying Slow Collective Variables in Biomolecular Dynamics

Area of Science:

• Computational Biology
• Biophysics
• Data Science

Background:

• Elucidating collective variables (CVs) is essential for understanding biomolecular dynamics.
• Existing methods like AMUSEt (Algorithm for Multiple Unknown Signals) for Koopman approximation face limitations with high-dimensional data due to memory constraints, requiring manual feature selection.
• This manual process is challenging for complex biological systems.

Purpose of the Study:

• To develop a novel method, AMUSET-TICA (AMUSEt-based Time-lagged Independent Component Analysis), for identifying slow CVs in biomolecular dynamics.
• To overcome the limitations of manual feature selection in previous methods.
• To provide a computationally efficient and accurate approach for analyzing complex biomolecular systems.

Main Methods:

• AMUSET-TICA utilizes time-structure-independent components (tICs) as input features for the AMUSEt algorithm.
• It embeds high-dimensional protein conformations by expanding orthogonal tICs into overlapping Gaussian basis functions via a tensor-product data structure.
• This approach avoids the need for manual feature selection and ranking.

Main Results:

• AMUSET-TICA significantly outperforms AMUSEt and tICA in identifying slow CVs across three test systems: alanine dipeptide, NTL9, and FIP35 WW domain.
• The identified CVs accurately describe the slowest dynamical modes of these systems.
• AMUSET-TICA demonstrates performance comparable to deep-learning methods like VAMPnets but with greater computational efficiency.
• The method provides mechanistic insights into protein folding, including parallel pathways for the FIP35 WW domain.

Conclusions:

• AMUSET-TICA is a robust and efficient method for identifying collective variables in biomolecular dynamics.
• It effectively handles high-dimensional data without manual feature engineering.
• The approach offers valuable insights into complex biological processes and protein folding mechanisms.
• AMUSET-TICA is expected to be widely applicable in biomolecular dynamics research.