(4lj) Molecular Science Discovery through Machine Learning-Based Forcefields and Electronic Structure Predictors

My research interests are rooted in both my academic and professional experiences, motivated by a commitment to merge provenance and prosperity. It centers around addressing two overarching challenges: first, the acceleration of atomistic simulations, and second, the translation of knowledge obtained from simulations into tangible experimental outcomes. Molecular simulations play a pivotal role in unraveling the nuances of molecular behaviors and material properties, offering invaluable insights into chemical reactions, material characteristics, and biological interactions. Information from such simulations is commonly used to perform high-throughput studies to discover novel molecules and materials with targeted properties. However, these simulations often encounter barriers related to accuracy, efficiency, and scalability, thereby constraining their reliability and applicability, especially when bridging the gap between simulation results and experimental validations. There is also a compromise to the extent of electronic information when dealing with large molecules and materials.

To address these challenges, my research objectives are geared towards integrating quantum chemistry, machine learning (ML)-based forcefields, and charge density predictors within a holistic active learning framework. This will provide an accelerated way to have geometric and electronic information about the system of interest. The primary focus is on expanding the chemical reactive space to enable comprehensive exploration and characterization of chemical reactions. This includes generating reactive training data without relying on predefined reaction pathways or products, thereby equipping ML forcefields with the capacity to encapsulate the full spectrum of chemical reactivity. Through an iterative active learning scheme, automatic generation of potential reactions coupled with transition-state finding methods facilitates the exploration of reactive space, strengthening the versatility and predictive accuracy of ML forcefields. Furthermore, this active learning approach extends to training ML models for 3-D charge density prediction, where quantum mechanics-based relabeling of new configurations is employed to generate corresponding electron density data. Ultimately, this iterative active learning process culminates in the development of accurate ML forcefields, and charge density predictors tailored to the specific system under investigation.

These ML forcefields will then be deployed for large-scale simulations, while the charge density predictors will be utilized to study charge transfer and perform chemical bond analysis. Integration with multi-objective high-throughput algorithms will enable the exploration of new molecules with targeted applications. By combining thermodynamic, electronic, and spectroscopic information generated from these models, novel molecules with tailored properties can be discovered, fostering innovation across various domains such as materials science, drug discovery, and energy storage.

Teaching Interests

My teaching philosophy is based on the idea that learning is an active and collaborative process that requires both guidance and challenge from the instructor. I am a strong believer in the idea that scientific progress is limited by an individual’s curiosity and care towards a topic, which is what controls their intelligence. I believe that students learn best when they are engaged in meaningful tasks that connect theory with practice, and when they receive constructive feedback that helps them improve their understanding and skills. As a teacher, I aim to create a supportive and inclusive learning environment that fosters curiosity, creativity, and critical thinking among students. As a Teaching Assistant for the graduate-level course “Fundamentals of Reaction Processes” at the University of Pittsburgh, I had the opportunity to deliver three lectures on topics related to my research interests: modeling chemical reactors using numerical solvers, computational chemistry, and molecular dynamics simulations. I designed these lectures to be interactive and application-oriented, using examples from real-world problems and current research. I received positive feedback from both the instructor and the students, who appreciated my clear explanations, relevant examples, and helpful suggestions. The subjects that I would like to teach are: Thermodynamics, Statistical Mechanics, Statistics and Machine Learning, Computational Chemistry, and Reaction Engineering.