Research

Developing Better Ways to Do Chemistry

Chemistry is challenging. Irreproducibility, Inefficiencies, and Inaccessibility make it even harder. We develop manually operated and automated flow chemistry platforms to minimize error and maximize reproducibility and output.

 

Single-Step systems: The development and study of new organic transformations, for in-depth mechanistic studies, and database generation for machine learning.

 

Single- and Multistep Systems: Our radial synthesizer (shown right) is a fully automated, remotely accessible flow chemistry platform capable of performing any single- or multistep flow chemical process. It has a number of unique attributes due to the sequential, non-simultaneous nature of the system that make the instrument a powerful tool for process development, screening and optimization, as well as for data generation.

  • Concept of radial synthesis, where flow reaction modules are arranged radially, and equally accessibly, about a core switching station

Developing Ways to Better Understand Chemistry

Trying to discern the relationship between the multitude of factors influencing the outcome of a chemical reaction - whether yield or selectivity - is incredibly challenging. Its even harder with only an empirical understanding. We use computational chemistry and machine learning to quantify influential parameters  and their influence to predict reaction conditions and outcomes.

Computational Chemistry: Identify and quantify steric and electronic factors that influence reactivity, map out the relative positions of molecules in multidimensional space, and use this understanding to design new reagents, reactions, and reaction conditions.

 

Machine Learning: Develop and train prediction algorithms for reaction outcome, prediction of reaction conditions to maximize selective outcome, and transfer this knowledge to related chemical transformations while gaining fundamental understanding of the principles guiding reactivity of reactions studied.

  • graphical abstract for predicting stereoselectivity using machine learning