Deep Learning in Computational Chemistry

ANAKIN-ME: Accurate NeurAl networK engINe for Molecular Energies

Deep neural network (NN) trained on quantum mechanical (QM) DFT calculations can learn an accurate and transferable potential for organic molecules.

ANI is a new method designed with the intent of developing transferable neural network potentials that utilize a highly-modified version of atomic environment vectors (AEV) as a molecular representation. AEVs provide the ability to train neural networks to data that spans both configurational and conformational space.

We utilized ANI to build a potential called ANI-1, which was trained on a subset of the GDB databases with up to 8 heavy atoms in order to predict total energies for organic molecules containing four atom types: H, C, N, and O. To obtain an accelerated but physically relevant sampling of molecular potential surfaces, we also proposed a Normal Mode Sampling (NMS) method for generating molecular conformations. Through a series of case studies, we show that ANI-1 is chemically accurate compared to reference DFT calculations on much larger molecular systems (up to 54 atoms) than those included in the training data set.

Effect of pH on Biomolecules

We are interested in developing methods to study how proteins and nucleic acids behave at different values of solvent pH. Enzyme activity is often strongly dependent on the pH of its environment, so it is important to develop a model that can capture pH effects on these biological systems.
Solvent pH affects the behavior of proteins and nucleic acids by adjusting the protonation state equilibria of certain ionizable residues (e.g., glutamate, aspartate, and lysine). We are improving constant pH molecular dynamics methodologies to enable us to study more complex biological systems than our current methods are capable of treating.

Oxygen Diffusion in a Cofactor-Independent Oxygenase

Small gas molecules like dioxygen (O2) are key in different enzymatic reactions in living organisms. In spite of extensive research effort, many aspects of O2 biocatalysis remain poorly understood. In particular, some recent work has focused on studying how O2 reaches the active site of oxygen-using enzymes. While earlier studies suggested that O2 diffused passively though proteins, recently it has been proposed that O2 follows highly-specific tunnels. Molecular Dynamics simulations, in combination with mutagenesis and kinetic studies, have been a fundamental tool in establishing the existence of multiple pathways for O2 diffusion into myoglobin and oxygenases that rely on metal cofactors or flavins. While the latter are vastly the most common type of oxygenases, several cofactor-independent oxidases and oxygenases have been identified.5 The detailed mechanism of how these oxygenases are able to activate triplet O2 while avoiding oxidative inactivation has not been established.
Our work focuses on vancomycin-biosynthetic dioxygenase DpgC, a cofactor-independent oxygenase for which a crystal structure in which O2 is observed is available. We are using molecular dynamics to study O2 diffusion pathways into its active site. We will support our computational findings using mutagenesis and kinetic studies that we will perform in collaboration with the Bruner group at the University of Florida. Future work will focus on gaining more insight into the mechanism of this enzyme.

Replica Exchange Methods

Our research focuses on tuning and developing new methods in free energy calculations, especially those that are based on Hamiltonian Replica Exchange Molecular Dynamics (HREMD). We are trying to combine HREMD with other conventional free energy calculation methods (e.g., Umbrella Sampling MD, Constant pH MD etc.) to increase the efficiency of sampling in phase space.

QM/MM Umbrella Sampling Replica Exchange

To combine the advantage of efficient sampling in Replica Exchange MD and Biased Sampling MD with the accuracy of QMMM methods, a QMMM Umbrella Sampling Replica Exchange (QMMM-USRE) method has been proposed. For justification of the benefits of this method and finding the role of exchange frequency, the free energy profile for the SN2 reaction between methyl chloride and a chloride has been computed using the QMMM-USRE method at different exchange rates.

Exchange Molecular Dynamics In Proteins Using a Discrete Protonation Method

We are developing a pH-Replica Exchange Molecular Dynamics (pH-REMD) method to improve the coupling between conformational and protonation sampling. This method is a combination of Constant pH MD and Hamiltonian Replica Exchange methods. Under a Hamiltonian replica exchange setup, conformations are swapped between two neighboring replicas, which themselves are at different pHs.