A gradient-directed Monte Carlo method for global optimization in a discrete space: application to protein sequence...

Area of Science:

• Computational Biology
• Biophysics
• Protein Design

Background:

• Selecting optimal protein sequences from a vast chemical space is computationally challenging.
• Conventional Monte Carlo methods struggle with discrete variables and navigating complex energy landscapes.
• Accurate calculation of local property gradients is crucial for efficient optimization.

Purpose of the Study:

• To introduce and evaluate the Gradient-Directed Monte Carlo (GDMC) method for protein sequence design and folding.
• To demonstrate GDMC's ability to efficiently explore discrete chemical spaces.
• To compare GDMC's performance against conventional Monte Carlo approaches.

Main Methods:

• Applied GDMC, a novel optimization technique, to a discrete space of chemically viable proteins.
• Utilized local property gradients derived from interpolated discrete property values (Linear Combination of Atomic Potentials).
• Incorporated the Metropolis criterion to overcome energy barriers and identify global minima.

Main Results:

• GDMC successfully selected optimal protein members within the discrete chemical space.
• The method demonstrated efficiency in protein sequence design and folding simulations using the HP lattice model.
• Local property derivative information effectively guided the search towards global energy minima.

Conclusions:

• GDMC is a highly efficient algorithm for optimizing discrete spaces, particularly in protein design and folding.
• The strategy shows significant promise for tackling complex inverse molecular design problems.
• GDMC's adaptability suggests broader applicability to various discrete optimization challenges.