(2bm) Advancing Crystallization to Enable Challenging Separations

Crystallization is the preeminent method for isolating molecules when purity is of utmost importance. Design of new crystal products (e.g. cocrystals, salts, solvates, complexes, etc., from here on collectively referred to as cocrystals) offers an avenue to improve products and processability of target molecules while maintaining desired purity. Recently, machine learning models have been developed that can predict whether two components will form a cocrystal with the same accuracy as traditional crystal structure prediction techniques but at a fraction of the computational cost. Likewise, models for predicting some properties of cocrystals, such as solubility, are becoming more accurate. However, our ability to exploit complex crystallization systems is still very much in its infancy. MD simulations are computationally expensive, data-driven models suffer from lack of diversity in training data, and many important crystal properties remain hard to predict. Additionally, discovering new cocrystals does not advance the development of improved products or processes. Can one design a coformer to optimize some property(ies) of the final cocrystal system? In answering this question, we hope to gain a greater understanding of cocrystallization as both a phenomenon and a process. My research group will leverage data-driven models to explore crystallization systems and will design separations processes to exploit those crystal products.

PhD Research (2015-2020):

I completed my PhD at Georgia Tech while advised by Ronald Rousseau, Martha Grover, and Andy Bommarius. Our research focused on coupling biocatalysis and crystallization in a continuous process. We were particularly interested in the synergistic effects of the coupled synthesis and separation; how in situ product removal by crystallization could enhance enzyme kinetics and precisely tuned enzyme kinetics could improve crystal shape and size distribution. Starting in my second year, we began working with the FDA to show that the process could be adapted to continuous manufacturing with improvements in yield, purity, and controllability. I gained expertise in crystallization (including cocrystallization, as we sought further process enhancements), kinetics, process analytical technology, and process control. We also developed new analytical techniques to probe some fundamentals of both crystallization and biocatalysis. These techniques enabled us to apply our findings to various beta-lactam antibiotics, develop several biocatalysts, and explore additional process intensification strategies.

Postdoc Research (2020-present):

I am presently working with Klavs Jensen at MIT as a postdoctoral researcher focused on developing an automated platform for the discovery and optimization of new molecules. Our work tests machine learning predictions to deliver improved models and ultimately discover molecules with finely tuned properties. So far, we have investigated dye molecules, aiming to improve their solubility, stability, and color, with limited human supervision. Our platform autonomously attempts to find optimal molecules by iteratively generating new molecules, predicting their properties, predicting synthetic routes to the molecules, synthesizing the molecules, characterizing the molecules’ properties, retraining the models with the results of the synthesis and characterization, and repeating until a top performer is discovered. I have contributed to the autonomous execution of multistep retrosynthetic predictions, automated characterization of several chemical properties, development of models for new properties, and implementation of active learning techniques to maximize model improvement.

Future directions

The skills I have acquired during my postdoc, along with the expertise in crystallization from my PhD, enable me to build a research group focused on answering the many outstanding questions in crystal design and processing. The space I intend to enter is underrepresented in US academia, international colleagues and industrial partners have made the most significant recent contributions. By cultivating expertise, with help from the pharmaceutical industry, in an academic environment, separation processes based on new crystallization technologies can be explored for more far-reaching challenges, such as water security, carbon capture, and green chemistry. Specifically, my group will:

1. Develop a platform for high-throughput screening of more diverse coformers and cocrystals

Cocrystal design suffers from a lack of training data and diversity. Machine learning models have begun to compete with traditional crystal structure prediction but at a fraction of the computation cost. In the Cambridge Structural Database (CSD) 10% of coformers are featured in 70% of cocrystals; the typical cocrystal discovery workflow involves screening targets against a panel of known cocrystal coformers. Matching hydrogen bonding patterns between target and coformer has been cited as a heuristic for designing cocrystals, however many interesting examples of cocrystals do not follow this rule. We will develop tools to screen new cocrystals with solutions-based techniques, allowing parallelization and access to a flexible design space. Uncertainty analysis and chemical diversity will maximize prediction improvement per experiment.

2. Design and improve on models of crystal and cocrystal properties, especially those pertaining to processability

Crystal product quality depends on properties of the molecule(s) that constitute the crystal lattice as well as properties of the lattice itself. Recent work has demonstrated the ability to predict robustly a variety of molecular properties with machine learning methods, however relevant properties of crystalline materials have remained a more challenging task. Properties such as crystal habit, hygroscopicity, solubility, and powder flowability are all critical to pharmaceutical development, yet we are only starting to be able to predict them for arbitrary products. My group will focus on using hierarchical representations of inter- and intramolecular interactions to build new models that capture molecular and lattice contributions to these complex properties. We will use combined first principles and data-driven approaches to enable robust predictions across a variety of compound families.

3. Apply bespoke crystal products to challenging separations and processes

One of the big questions my group will seek to answer is, given a target molecule (or chemical entity) and a property, can one design a crystal product to optimize said property? Answering this question requires designing a crystallization process as well as potential coformers/additives to modify the crystal product. Process metrics such as yield are equally important optimization targets, especially when considering difficult separations such as removal of dilute contaminants or chemically similar species. Specific applications include design of chiral resolving agents, recovery of valuable metals from seawater, separation of benzene derivatives, and removal of inhibitory fermentation products.

Teaching Interests:

Chemical Engineering Core Courses: Mass and Energy Balances, Separations, Kinetics and Reactor Design

Chemical Engineering Electives: Crystallization science, Pharmaceutical Discovery and Manufacturing, Mathematical Modeling in Chemical Engineering

The attention paid to teaching by my undergraduate professors is part of what led me to pursue research in chemical engineering. I believe that teaching is an opportunity to expose future researchers to the cutting edge of chemical science. I plan to pepper lectures with examples relevant to challenges we as chemical engineers are well suited to solve, as well as relevant examples from my research and my colleagues’ research. My approach to teaching aims to maximize engagement with the material so that clearly defined learning checkpoints are realized in pursuit of answering larger questions and satisfying broader curiosities. Even the most advanced research begins with the fundamentals, making this approach applicable to introductory and graduate-level courses. During both my PhD and postdoc, I have mentored several undergraduates and found the experience to be very rewarding in terms of understanding how other approach problems; motivation to solve a problem inspires learning. Building interest in the power of a STEM education goes beyond the classroom, I plan to continue outreach with the local community to demonstrate how chemical science can explain seemingly complex and amusing phenomena.