Publications

2024

1. 

   State-specific protein–ligand complex structure prediction with a multiscale deep generative model

   Qiao, Zhuoran, Nie, Weili, Vahdat, Arash, Miller III, Thomas F, and Anandkumar, Animashree

   Nature Machine Intelligence 2024

2023

1. COVIDisAirborne: AI-enabled multiscale computational microscopy of delta SARS-CoV-2 in a respiratory aerosol

   Dommer, Abigail, Casalino, Lorenzo, Kearns, Fiona, Rosenfeld, Mia, Wauer, Nicholas, Ahn, Surl-Hee, Russo, John, Oliveira, Sofia, Morris, Clare, Bogetti, Anthony, and others,

   The international journal of high performance computing applications 2023

2022

1. Multi-modal Molecule Structure-text Model for Text-based Retrieval and Editing

   Liu, Shengchao, Nie, Weili, Wang, Chengpeng, Lu, Jiarui, Qiao, Zhuoran, Liu, Ling, Tang, Jian, Xiao, Chaowei, and Anandkumar, Anima

   arXiv preprint arXiv:2212.10789 2022
2. Dynamic-Backbone Protein-Ligand Structure Prediction with Multiscale Generative Diffusion Models

   Qiao, Zhuoran, Nie, Weili, Vahdat, Arash, Miller III, Thomas F, and Anandkumar, Anima

   NeurIPS 2022 Machine Learning in Structural Biology (MLSB) Workshop, contributed talk. 2022

   HTML
3. Retrieval-based Controllable Molecule Generation

   Wang, Zichao, Nie, Weili, Qiao, Zhuoran, Xiao, Chaowei, Baraniuk, Richard, and Anandkumar, Anima

   arXiv preprint arXiv:2208.11126 2022

   HTML
4. 

   Informing geometric deep learning with electronic interactions to accelerate quantum chemistry

   Qiao, Zhuoran, Christensen, Anders S., Welborn, Matthew, Manby, Frederick R., Anandkumar, Anima, and Miller, Thomas F.

   Proceedings of the National Academy of Sciences 2022

   Abs HTML

   Predicting electronic energies, densities, and related chemical properties can facilitate the discovery of novel catalysts, medicines, and battery materials. However, existing machine learning techniques are challenged by the scarcity of training data when exploring unknown chemical spaces. We overcome this barrier by systematically incorporating knowledge of molecular electronic structure into deep learning. By developing a physics-inspired equivariant neural network, we introduce a method to learn molecular representations based on the electronic interactions among atomic orbitals. Our method, OrbNet-Equi, leverages efficient tight-binding simulations and learned mappings to recover high-fidelity physical quantities. OrbNet-Equi accurately models a wide spectrum of target properties while being several orders of magnitude faster than density functional theory. Despite only using training samples collected from readily available small-molecule libraries, OrbNet-Equi outperforms traditional semiempirical and machine learning–based methods on comprehensive downstream benchmarks that encompass diverse main-group chemical processes. Our method also describes interactions in challenging charge-transfer complexes and open-shell systems. We anticipate that the strategy presented here will help to expand opportunities for studies in chemistry and materials science, where the acquisition of experimental or reference training data is costly.

2021

1. OrbNet Denali: A machine learning potential for biological and organic chemistry with semi-empirical cost and DFT accuracy

   Christensen, Anders S., Sirumalla, Sai Krishna, Qiao, Zhuoran, O’Connor, Michael B., Smith, Daniel G. A., Ding, Feizhi, Bygrave, Peter J., Anandkumar, Animashree, Welborn, Matthew, Manby, Frederick R., and Miller, Thomas F.

   The Journal of Chemical Physics 2021

2020

1. 

   Multi-task learning for electronic structure to predict and explore molecular potential energy surfaces

   Qiao, Zhuoran, Ding, Feizhi, Welborn, Matthew, Bygrave, Peter J, Anandkumar, Animashree, Manby, Frederick R, and Miller III, Thomas F

   arXiv preprint arXiv:2011.02680 2020 (Appeared at Machine Learning for Molecules workshop at NeurIPS 2020 as a contributed talk)

   Abs HTML

   We refine the OrbNet model to accurately predict energy, forces, and other response properties for molecules using a graph neural-network architecture based on features from low-cost approximated quantum operators in the symmetry-adapted atomic orbital basis. The model is end-to-end differentiable due to the derivation of analytic gradients for all electronic structure terms, and is shown to be transferable across chemical space due to the use of domain-specific features. The learning efficiency is improved by incorporating physically motivated constraints on the electronic structure through multi-task learning. The model outperforms existing methods on energy prediction tasks for the QM9 dataset and for molecular geometry optimizations on conformer datasets, at a computational cost that is thousand-fold or more reduced compared to conventional quantum-chemistry calculations (such as density functional theory) that offer similar accuracy.

2. 

   OrbNet: Deep learning for quantum chemistry using symmetry-adapted atomic-orbital features

   Qiao, Zhuoran, Welborn, Matthew, Anandkumar, Animashree, Manby, Frederick R, and Miller III, Thomas F

   The Journal of Chemical Physics 2020 (Editor’s Pick)

   Abs HTML

   We introduce a machine learning method in which energy solutions from the Schrödinger equation are predicted using symmetry adapted atomic orbital features and a graph neural-network architecture. OrbNet is shown to outperform existing methods in terms of learning efficiency and transferability for the prediction of density functional theory results while employing low-cost features that are obtained from semi-empirical electronic structure calculations. For applications to datasets of drug-like molecules, including QM7b-T, QM9, GDB-13-T, DrugBank, and the conformer benchmark dataset of Folmsbee and Hutchison [Int. J. Quantum Chem. (published online) (2020)], OrbNet predicts energies within chemical accuracy of density functional theory at a computational cost that is 1000-fold or more reduced.

3. Systems and Methods for Determining Molecular Structures with Molecular-Orbital-Based Features

   Miller, Thomas F, Welborn, Matthew G, Cheng, Lixue, Husch, Tamara, Song, Jialin, Kovachki, Nikola, Burov, Dmitry, Teh, Ying Shi, Anandkumar, Anima, Ding, Feizhi, Lee, Sebastian, Qiao, Zhuoran, and Lale, Ali Sahin

   2020 US Patent App. 16/817,489

2019

1. 

   Ice nucleation of confined monolayer water conforms to classical nucleation theory

   Qiao, Zhuoran, Zhao, Yuheng, and Gao, Yi Qin

   The journal of physical chemistry letters 2019

   Abs HTML

   We confirmed that monolayer water confined by parallel graphene sheets spontaneously crystallizes from a structurally and dynamically heterogeneous liquid phase under moderate supercooling via direct molecular dynamics simulation. Square-lattice-like geometric order is observed at the early stage of nucleation and is preserved during the entire nucleus growth process. The diffusion coefficient and free energy profile in the cluster space extracted from a Bayesian trajectory analysis agree well with the classical nucleation theory (CNT) prediction and yield thermodynamic quantities exhibiting linear temperature dependence. The effectiveness of maximum cluster size as the descriptor of ice nucleation dynamics in the CNT framework can be attributed to the dynamical time scale decoupling and strong structural pattern dependence of density fluctuation in the liquid phase.

2. Structure of water confined between two parallel graphene plates

   Cai, Xiaoxia, Xie, Wen Jun, Yang, Ying, Long, Zhuoran, Zhang, Jun, Qiao, Zhuoran, Yang, Lijiang, and Gao, Yi Qin

   The Journal of chemical physics 2019

   Abs HTML

   We study, in this paper, the physical properties of water confined between two parallel graphene plates with different slit widths to understand the effects of confinement on the water structure and how bulk properties are reached as the water layer thickens. It was found that the microscopic structures of the interfacial liquid layer close to graphene vary with the slit width. Water tends to locate at the center of the six-membered ring of graphene planes to form triangular patterns, as found by others. The narrower the slit width is, the more pronounced this pattern is, except for the slit width of 9.5 Å, for which a well-defined two-layer structure of water forms. On the other hand, squared structures can be clearly seen in single snapshots at small (6.5 Å and 7.5 Å) but not large slit widths. Even at small slit widths, the square-like geometry is observed only when an average is taken for a short trajectory, and averaging over a long time yields a triangular pattern dictated by the graphene geometry. We estimate the length of time needed to observe two patterns, respectively. We also used the two-phase thermodynamic model to study the variation of entropy of confined water and found that at 8.5 Å, the entropy of confined water is larger than that of bulk water. The rotational entropy of confined water is higher than that of bulk water for all slit widths due to the reduction of the hydrogen bond in the confined space.

3. Interlayer hopping dynamics of bilayer water confined between graphene sheets

   Qiao, Zhuoran, Xie, Wen Jun, Cai, Xiaoxia, and Gao, Yi Qin

   Chemical Physics Letters 2019

   Abs HTML

   The ubiquitous existence of water confined in nano-capillaries is key to fundamental biological and technological applications. Using molecular dynamics simulations, we analyzed the hopping-like interlayer relocation dynamics of bilayer water confined between two parallel graphene sheets. In contrary to the common scheme that relocation is driven by density fluctuations, analysis of the transition path ensemble revealed that interlayer hopping is induced by local hydrogen bond configuration fluctuations coupled with activated consecutive transient angular reorientations. Our results consolidated the anisotropic nature of water relocation under strong ordering, which provided a mechanistic insight into the relaxation behavior at glass-forming water’s fragile-to-strong crossover.