ConfRank+: Extending Conformer Ranking to Charged Molecules.

Area of Science:

• Computational Chemistry
• Machine Learning
• Quantum Chemistry

Background:

• Accurate energetic ranking of molecular conformers is crucial for computational chemistry.
• Existing methods can be computationally expensive, limiting high-throughput applications.
• Machine learning offers a promising avenue for accelerating these predictions.

Purpose of the Study:

• To develop a machine learning model for high-throughput energetic ranking of charged molecular conformers.
• To create a multifidelity model capable of emulating multiple reference quantum chemistry methods.
• To incorporate molecular charge distribution information for improved accuracy.

Main Methods:

• A pairwise learning approach based on the ConfRank method was employed.
• The model was trained on two distinct reference levels simultaneously using dataset embedding vectors.
• Partial atomic charges from the electronegativity equilibration charge model were included to represent charge distribution.

Main Results:

• The multifidelity machine learning model accurately reproduces two different DFT reference levels for small- to medium-sized molecules.
• The model demonstrates comparable accuracy to state-of-the-art methods like AIMNet2 and MACE-OFF23(L).
• It requires an order of magnitude fewer parameters than existing models and matches the robustness of GFN2-xTB.

Conclusions:

• The developed machine learning model provides a computationally efficient and accurate solution for ranking charged molecular conformers.
• This approach enables high-throughput screening with high fidelity, advancing molecular modeling capabilities.
• The model's ability to handle charged species and electrostatic interactions broadens its applicability in chemical research.