A new method for the gradient-based optimization of molecular complexes.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Structural Biology

Background:

• Molecular complexes are often simplified using compact representations in molecular modeling.
• This simplification, while reducing degrees of freedom, introduces singularities problematic for gradient-based optimization.

Purpose of the Study:

• To introduce a novel method for local optimization of molecular complexes suitable for molecular docking.
• To address and overcome the singularities encountered in gradient-based optimization of compact molecular representations.

Main Methods:

• Employing exponential mapping for molecular orientation to simplify gradient calculation.
• Modifying local minimization algorithms to perform well-defined jumps, avoiding parametrization singularities.
• Applying the method to continuous, potentially non-differentiable objective functions.

Main Results:

• The novel method successfully optimized molecular complexes with varying internal degrees of freedom against large receptors.
• Demonstrated superior performance compared to the Solis and Wets method, especially with non-differentiable scoring functions.
• Achieved significantly improved scores in fewer optimization steps across all tested complexes.

Conclusions:

• The proposed method offers an effective solution for gradient-based optimization of compact molecular representations in molecular docking.
• This approach enhances computational efficiency and accuracy in molecular complex optimization.
• The technique provides a robust alternative for handling singularities in molecular structure optimization.