Advancements in mathematical methods and computing technology have revolutionized computational science and engineering, making it indispensable in chemistry. Computational research facilitates experimental study design, validation, and investigation of topics where experiments are impractical. In this context, our research focuses on developing new computational chemistry methodologies to efficiently model/simulate diverse molecular systems and accurately predict their chemical properties. We adopt quantum computing principles to create quantum machine learning models and design algorithms for quantum resource utility. Additionally, we are interested in training classical machine learning models with improved generalization and accuracy. We also develop techniques that enhance conventional quantum chemistry methods leading to physics-aware methodologies. Our primary objective is to develop effective computational tools that enable molecular modeling, structure-function analysis, and high-throughput screening. We are interested in creating a data-driven and digital chemistry approach to impact applications areas such as reaction design, catalyst discovery, and drug development.
​
Quantum Computing
Our research focuses on advancing chemistry through the application of quantum computing principles, particularly by developing quantum machine learning (QML) models that distribute workloads between classical and quantum computers. We are exploring the feasibility of quantum enhancement in building such QML models for various tasks, aiming to understand how quantum computing offers new ways to process data and build more efficient models with less data. Our work investigates the potential improvements QML can bring in predicting molecular properties and generating chemical data, while also developing innovative search strategies to discover better-performing models. Additionally, we are focused on quantum resource optimization, as current quantum computers are still in their early stages and face limitations, including limited processing capacity and susceptibility to errors. To address these challenges, we are interested in creating techniques and algorithms that optimize quantum devices for chemical applications. Our efforts include minimizing quantum processing requirements, accelerating parameter tuning, enhancing control through adjustable parameters, implementing error mitigation strategies, distributing workloads across multiple quantum devices, and improving task distribution between classical and quantum resources.
By combining machine learning algorithms with computational chemistry, our work involves developing models that can provide fast and accurate molecular property predictions, overcoming the computational challenges of conventional quantum mechanical (QM) methods. While machine learning reduces computational costs, it often faces issues with poor generalization and limited training data. To address these challenges, we are interested in integrating QM information into the training step, developing correction models for less accurate QM methods, and creating large, high-quality datasets using in-house approaches to improve model performance. These efforts can enable rapid, accurate predictions of chemical properties for various applications. For instance, we are developing machine learning models to predict bond dissociation enthalpies, which is crucial for determining chemical bond strengths and providing insights into molecular stability, facilitating the design of organic antioxidants, and identification of drug autooxidation sites. We are also developing a machine learning-driven synthesis planning tool aimed at identifying cost-effective and sustainable synthetic routes, reducing the experimental reliance on trial-and-error. Additionally, we are developing models to predict transition states—key structures in chemical reactions—much faster and more efficiently than traditional methods.
Machine Learning
Quantum Chemistry
Our research in quantum chemistry involves advancing the accuracy and efficiency of electronic structure methods for performing numerous computations in a short period or handling large molecular systems. By developing and optimizing atom-centered potentials (ACP), we bridge the gap between computationally inexpensive quantum mechanical methods and the high-level accuracy required for high-throughput applications. ACP can correct the limitations of small-basis-set Hartree–Fock (HF) and density functional theory (DFT) methods, enabling accurate predictions of molecular properties such as non-covalent interaction energies, reaction barriers, bond dissociation energies, and molecular geometries. This approach supports the efficient modeling of complex molecular systems with hundreds of atoms while retaining computational feasibility. Additionally, we integrate ACP with semi-empirical corrections like D3 to enhance the accuracy and applicability of HF and DFT methods for biochemical and organic chemistry applications. By generating large, high-quality datasets through ACP-driven workflows, we also advance deep learning models for quantum chemistry, creating tools that achieve near coupled-cluster level of theory accuracy at a reduced computational cost.
We harness the power of data for chemistry by creating diverse benchmark data sets of molecular properties, developing and undertaking use-case chemical applications (reaction discovery, catalyst optimization, covalent drug design,...), and providing computational screening support to experimental collaborations. Our goal is to leverage data-driven approaches to uncover new insights and accelerate discoveries, embracing a modern, digital approach to chemistry. By addressing the gaps in existing reference data sets, we have developed high-quality resources such as BH9, BSE49, and PEPCONF. These data sets provide consistent benchmarks for barrier heights, reaction energies, bond dissociation reactions, and peptide conformers, setting new standards for evaluating or developing computational methodologies. They eliminate the variability and uncertainty of data from disparate sources, offering researchers reliable tools to assess and refine their methods. In experimental collaborations, we are investigating the reaction energetics of a large pool of candidates for a particular organic reaction type to identify novel synthetic targets. We are interested in developing machine learning models that leverage this data to enable high-throughput exploration of chemical reaction space.
​
Data-driven Chemistry
Funding
Resources
