Multi-dimensional spectral gap optimization of order parameters (SGOOP) through conditional probability factorization

Area of Science:

• Computational Chemistry
• Molecular Dynamics
• Statistical Mechanics

Background:

• Rare events in molecular systems are challenging to sample using traditional methods.
• Constructing accurate reaction coordinates (RC) is crucial for understanding molecular processes.
• Spectral Gap Optimization of Order Parameters (SGOOP) is a method for RC construction, originally limited to 1D.

Purpose of the Study:

• To extend SGOOP for constructing multi-dimensional reaction coordinates.
• To develop a method for identifying and learning hidden features of molecular energy landscapes.
• To improve the efficiency and applicability of SGOOP for complex molecular systems.

Main Methods:

• Introduced conditional probability factorization to SGOOP.
• Developed a sequential, bottom-up approach to probe and extend RC dimensionality.
• Applied the extended SGOOP to learn a 2D RC for benzene unbinding from T4L99A lysozyme.

Main Results:

• Demonstrated systematic probing for RC dimensionality extension.
• Successfully learned additional RC coordinates in a computationally efficient manner.
• Obtained excellent agreement for benzene unbinding kinetics and pathway with experiments and other methods.

Conclusions:

• The extended SGOOP method enables the construction of multi-dimensional RCs for complex systems.
• This approach facilitates intuition-free sampling and enhances the utility of SGOOP and other RC construction methods.
• The method is effective for challenging problems like ligand unbinding kinetics.