(2bj) Machine Learning-Assisted Materials Design for Energy and Sustainability
AIChE Annual Meeting
2022
2022 Annual Meeting
Meet the Candidates Poster Sessions
Meet the Faculty and Post-Doc Candidates Poster Session
Sunday, November 13, 2022 - 1:00pm to 3:00pm
In Basdogan Lab we will combine molecular simulations, quantum chemistry calculations, and artificial intelligence to design new materials for sustainable energy applications and study systems that will help us have a fundamental understanding of environmental problems. My previous research focused on studying the essentials of solvated processes and developing efficient mixed explicit/continuum solvation procedures to correct deficiencies in continuum solvation models. Currently, I have been working on investigating polymer membranes for reducing greenhouse gas emissions by Machine Learning (ML) assisted polymer engineering, molecular dynamics simulations, and high throughput screening. With my experience using molecular dynamics simulations, along with first-principles electronic structure calculations utilizing density functional theory (DFT) and the knowledge of polymer materials and separation processes and various ML techniques, I propose several research areas in which my expertise and knowledge will be leveraged.
Aim 1: Designing new polymer materials for various applications like membrane separation processes, and polyelectrolytes for lithium battery systems. The two main problems that we have been facing with our current Machine Learning project are:
(1) Relying on an experimental dataset limits the power of machine learning models since we do not have any control with our input dataset.
(2) Lack of appropriate methods or standards for converting complicated systems like polyelectrolytes or copolymers into chemically informed, machine-readable representations.
In my independent research, I would like to study these systems in detail with molecular dynamics simulations and create a dataset where we have a full control. I would like to use simulation metadata by formulating vectors of force-field parameters that are specific to each repeating unit in the polymer. I believe we can learn from every simulation while we create the dataset and capture the physics of the polymer for our ML models since force field parameters express information such as the repeating unit size or its interaction with other moieties, they are somewhat like common descriptors like accessible surface area, partitioning coefficients, or properties derived from quantum chemical calculations.
Aim 2: Studying ion solvation in mixed solvents and polyelectrolyte solvation structures in different systems. During my PhD studies, we have looked at number of different ions in water and investigated their solvation structures using mixed explicit/continuum solvation procedures and ML. We had great success identifying the first solvation shell and solvation free energies for monovalent and divalent ions. Using my previous experience, I have plan to investigate two areas where there is still a need for a molecular level understanding and characterization for two specific systems:
(1) Ion solvation in mixed solvents such as: water/methanol and water/acetonitrile mixtures.
(2) Structure properties of polyelectrolytes in different solutions
Moving forward, I believe first principal study on systems stated above is necessary. Only quantum level calculations can help us identify the true physical structure and the properties of interest. We will create low energy molecular clusters with different numbers of explicit solvent molecules and calculate their properties using high level quantum chemistry calculations. Furthermore, we will study these clusters with Smooth Overlap of Atomic Positions (SOAP) kernel to quantify the similarity between different low-energy solvent environments. This unsupervised ML approach will help us identify the solvation shell of these complicated systems.
Aim 3: Investigating reaction mechanisms on electrocatalytic and photocatalytic CO2 reduction. Even though there has been a lot of interest in CO2 chemistry in the past decade, there is still a lack of fundamental molecular level understanding of the reaction mechanisms. Both electrocatalytic and photocatalytic CO2 reduction are a promising routes to efficiently convert CO2 into more useful products but there are shortcomings in two main areas:
(1) Identifying transition state structures and calculating reaction barrier heights
(2) Finding efficient cluster representations of the systems such that we can capture all the physical aspects as well as performing high level quantum chemistry calculations for barrier heights.
My students will perform static quantum chemistry modeling treatments to study how solvent molecules affect chemical reaction mechanisms without dynamics simulations. Computationally modeling atomic scale chemical reaction mechanisms in solvents is often not trivial. The most reliable and robust schemes usually involve dynamics-based treatments with explicit solvation models like metadynamics, transition path sampling, or umbrella sampling schemes that involve large numbers of electronic structure calculations. While such efforts can be very insightful, they can also bring very large computational costs and/or technical challenges that restrict their use. Our group will have a unique opportunity to study these system with ab-initio calculations while identifying the transition state structures and the barrier heights.
Research Experience
Currently, I am a postdoctoral researcher working with Dr. Zhen-Gang Wang at The California Institute of Technology. My first project focused on developing Simulation Models and Simulate Gas Permeation in Polyethers and MOPs. We provide a complete picture of how ether oxygen content in polyethers affects the membrane performance by focusing on five different polymers with increasing O:C ratio. These polymers are; PE: polyethylene (O:C=0), PTMO: polytetramethylene oxide (O:C=0.25), PEO: polyethylene oxide (O:C=0.50), PDXLA: poly(1,3-dioxolane) acrylate (O:C=0.67), and POM: polyoxymethylene (O:C=1.00). We can significantly control the solubility selectivity performance of a membrane material by increasing the oxygen ether content in the polymer. Furthermore, all polymers in our library have similar diffusivity selectivity, which slightly increases with the increasing O:C ratio. We note that the highest diffusivity selectivity all gas pairs are calculated in the POM membrane. Further research efforts focus on decorating the chain ends of polymer membranes with different functional groups to improve the performance.
My next project focuses on developing novel polymer membranes with high gas permeability and selectivity by ML assisted polymer engineering and high throughput screening. The polymer membrane materials space is vast, which includes all systems reported in the literature as well as the practically infinite polymer design space by altering the chemical groups in the molecular structure. We have put together a library of polymers from literature to study their structural properties that affects membrane selectivity and permeability. Elucidating these relationships enables accelerated polymer membrane design for a given gas separation application. To this end, we use a ML study of gas solubility and diffusivity in polymer membranes. Our central hypothesis is that ML models can accurately predict gas permeability and selectivity of both existing and hypothetical membranes when trained on large and diverse data.
My PhD research in Dr John Keithâs group at University of Pittsburgh focused on molecular level understanding and characterization of solvation environments. In many cases, the explicit interactions between molecules with nearby solvents are crucial for molecular-scale understanding. Toward practical modelling of local solvation effects of any solute in any solvent, we developed a general, all-QM, cluster-continuum approach. This approach uses a global optimization procedure to identify low energy molecular clusters with different numbers of explicit solvent molecules and then employs a machine learning algorithm with the help of the Smooth Overlap of Atomic Positions (SOAP) kernel to quantify the similarity between different low-energy solvent environments. From these data, we use a sketch-map non-linear dimensionality reduction technique to obtain a visual representation of the similarity between solvent environments in differently sized microsolvated clusters. After studying the evolution of the local solvation environment around the molecules, we systematically explore reaction pathways using Growing String Method. Without needing either dynamics simulations or an a priori knowledge of the local solvation structure, this procedure was used to calculate reaction energies, solvation free energies and barrier heights in solvated systems. We now use this approach to model reaction mechanisms in more complicated reaction environments that are relevant for renewable fuels and chemicals. We reliably predict CO2 hydrogenation pathways and calculate barrier heights under electrochemical environments. This approach can be used to study physically significant solvation environments in any solvated system where the solvent molecules affect the quantum level nature of reaction mechanisms.
Teaching Interests
My interest in pursuing an academic career was instigated by my love of teaching and mentoring. Helping students grow and develop their passion for research to the point where they are excited to work independently and propose new ideas has been extremely rewarding. My teaching and research mentorship experiences have taught me an important lesson: while it is rewarding to produce good science, my biggest impact will come from producing great scientists. Furthermore, I strongly believe people with diverse personal, ethnic, cultural, religious, and academic backgrounds can make STEM stronger therefore I am committed to creating a diverse, inclusive, and accessible culture in the classroom and my research group. I will actively use my leadership role as a professor to advocate for the support of underrepresented minorities in STEM, both within my university and in the broader community.
My background in chemical engineering has prepared me to teach any core course in the undergraduate or graduate curriculum. Furthermore, I am interested in developing (i) an undergraduate-level course introducing the basic techniques of molecular modeling for solving problems in chemical engineering, and (ii) a course for graduate students or advanced undergraduates where the students will be introduced to qualitative quantum chemistry concepts.