Molecular Conformer Search with Low-Energy Latent Space

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Molecular Modeling

Background:

• Accurately identifying low-energy molecular conformers is crucial but challenging for molecules with many degrees of freedom.
• Current methods often struggle with computational cost and efficiency for large molecular systems.

Purpose of the Study:

• To develop an efficient and accurate method for molecular conformer searching.
• To leverage generative models for exploring the low-energy potential energy surface of molecules.

Main Methods:

• Utilized molecular dihedral angles as features for a variational auto-encoder (VAE).
• Biased the VAE towards low-energy configurations to generate informative training data.
• Built a reliable energy model on the learned low-energy potential energy surface.
• Extracted and refined local-minimum conformations using structure optimization.

Main Results:

• Developed and benchmarked the low-energy latent-space (LOLS) structure search method.
• Successfully applied the LOLS method to organic molecules with 5-9 searching dimensions.
• Achieved results consistent with previous established studies, validating the approach.

Conclusions:

• The LOLS method provides an effective strategy for accurate molecular conformer searching.
• Using VAEs in a biased latent space can significantly improve the efficiency of identifying low-energy molecular configurations.
• This approach holds promise for advancing computational chemistry and molecular modeling accuracy.