MoleculeMind announced on September 14 a paper in which QuantaMind simulated a PETase enzyme reaction in a 17,792-atom system
MoleculeMind announced on September 14 a paper in which QuantaMind simulated a PETase enzyme reaction in a 17,792-atom system
The paper in Science Advances was published on September 11. On September 14, MoleculeMind announced the work in which PETase cleaves a PET fragment in a simulation that includes the protein, the substrate, and 4,898 water molecules.
Molecular dynamics is a calculation that moves atoms step by step under interatomic forces. The result is a reaction trajectory: a bond can form or break, and a proton can transfer to a neighboring atom.
Conventional force fields take these forces from predefined rules and can generate long trajectories. Quantum-chemical calculations account for electrons and therefore describe bond rearrangement, but for a large biomolecule they require heavy computation. In hybrid methods, the region where the reaction takes place is treated quantum-mechanically while the surroundings are handled with a simpler model; the boundary is chosen in advance.
QuantaMind is a neural-network force field: given atomic coordinates, it predicts the energy and forces for the next step of the trajectory. The training set included quantum-chemical calculations and configurations near transition states, which are rare atomic arrangements where one bond breaks and another forms.
PETase is a bacterial enzyme that breaks down PET plastic. The authors chose it because crystal structures are available for this enzyme and because the proton transfers in its catalytic mechanism remained a matter of debate. In the simulation, the serine residue in the active site attacked the substrate bond, histidine accepted and returned a proton, and water participated in product release and restoration of the active site.
At selected frames along the trajectory, the authors recalculated the forces with a quantum-chemical method; the correlation coefficient between these forces and the model's predictions exceeded 0.99. This comparison tests the forces specifically in the configurations that the model visits during the reaction.
For an unfamiliar molecule, the authors propose augmenting the training set with fragments of its trajectory: energies and forces are calculated for those fragments, then added to the model. In the case of PETase, 226 configurations were added in this way. This cycle makes it possible to refine the calculation for a specific reaction using the reaction's own configurations.
Source code and a short demonstration are published in the MoleculeMind repository; academic researchers can obtain the model parameters through a noncommercial request.