Development of neural network force fields (NNFFs) for exploration of enzymatic chemical reactions

An intense research efforts in the last few years have focused on the generation of machine-learning force fields for biological molecules, such as SpookyNet or AI2BMD. These offer the accuracy of quantum-chemical methods with many orders of magnitude speedups. However, these models were mostly focused on replacing traditional MM force fields, so they were less concerned about being accurate in the regime of chemical reactions where bonds are formed and broken. On the other hand, some models, such as OrbNet or AIMNet2 were explicitly designed to be universal. The aim of this project will be to test if, by generating extensive amounts of judiciously chosen training data, it would be possible to generate NNFFs that could speed up exploration of enzymatic reaction landscapes. Building upon the PeptideCS dataset, which is and exhaustive mapping of non-reactive potential energy landscape of peptides, your task will be to come up with suitable training data for, at first, a particular enzymatic reaction, such as ester bond hydrolysis, and later studying the universality-accuracy Pareto front by including more groups of reactions at once, eventually arriving at "universal" model. This will involve designing suitable benchmark problems, running reference calculation by DFT, large-scale generation of training data, training and modifying published NN models on custom datasets, and evaluation of their performance. The ultimate goal will be to remove the computational time bottleneck in the exploration of enzymatic reactions.

Field of study: Modelling of chemical properties on nano- and biostructures

References:

• Unke, O. T.; Chmiela, S.; Sauceda, H. E.; Gastegger, M.; Poltavsky, I.; Schütt, K. T.; Tkatchenko, A.; Müller, K. Machine Learning Force Fields. Chem. Rev. 2021, 121, 10142-10186. doi:10.1021/acs.chemrev.0c01111
• Wang, Y.; Wang, T.; Li, S.; He, X.; Li, M.; Wang, Z.; Zheng, N.; Shao, B.; Liu, T. Enhancing geometric representations for molecules with equivariant vector-scalar interactive message passing. Nat. Commun. 2024, 15, 313. doi:10.1038/s41467-023-43720-2
• Frank, J. T.; Unke, O. T.; Müller, K.; Chmiela, S. A Euclidean transformer for fast and stable machine learned force fields. Nat. Commun. 2024, 15, 6539. doi:10.1038/s41467-024-50620-6
• Anstine, D.; Zubatyuk, R.; Isayev, O. AIMNet2: A Neural Network Potential to Meet your Neutral, Charged, Organic, and Elemental-Organic Needs. ChemRxiv 2024. doi:10.26434/chemrxiv-2023-296ch-v3
• Zhang, S.; Makoś, M. Z.; Jadrich, R. B.; Kraka, E.; Barros, K.; Nebgen, B. T.; Tretiak, S.; Isayev, O.; Lubbers, N.; Messerly, R. A.; Smith, J. S. Exploring the frontiers of condensed-phase chemistry with a general reactive machine learning potential. Nat. Chem. 2024, 16, 727-734. doi:10.1038/s41557-023-01427-3