Eliminating the Deadwood: A Machine Learning Model for CCS Knowledge-Based Conformational Focusing for Lipids

Area of Science:

• Computational Chemistry
• Analytical Chemistry
• Machine Learning

Background:

• Accurate gas-phase chemical structure elucidation relies on combining experimental data with in silico predictions.
• Flexible molecules like fatty acids present challenges due to vast conformational spaces and high computational costs for sampling.

Purpose of the Study:

• To develop a machine learning model for efficient conformer filtering based on estimated gas-phase collision cross section (CCS) values.
• To reduce computational expenses in predicting structures of flexible lipid molecules.

Main Methods:

• A novel machine learning (ML) model was developed to estimate gas-phase CCS values for conformer filtering.
• The ML model was integrated into a computational workflow for conformational sampling of lipids.
• Quantum mechanics optimizations were used to evaluate the energy of selected conformations.

Main Results:

• The ML-driven approach achieved an average CCS prediction error of approximately 2% for lipid systems.
• Conformations selected using CCS focusing generally resulted in lower energy geometries compared to unfiltered sets.
• The model improved both the quality of structure prediction and the overall processing time.

Conclusions:

• The implemented ML model significantly enhances the accuracy and efficiency of gas-phase structure prediction for lipids.
• This CCS knowledge-based approach offers a valuable tool for computational chemistry, with potential for extension to other molecular classes.