Introduction

Optimization algorithms experience widespread application in economics, molecular modeling, and industrial processes. In chemistry, optimization of reactions for objectives like desired product yield, selectivity, cost efficiency, and conversion is a necessity as well. This is typically performed using a chemist’s intuition combined with one variable at a time (OVAT) or Design of Experiments (DoE) techniques. Mathematical optimization algorithms designed in other fields have found success when applied to chemical reaction optimization, but the various interfaces, programming skills, and technical debugging, and installation methods required to use an optimization algorithm is usually not considered worth the effort by most chemists. When targeting reaction optimization, the algorithms are expected to be used in the laboratory, likely by someone without much programming experience. Compounded on these issues, there is no one-size-fits-all algorithm so chemists need to switch algorithms based on which will perform the best on their reaction of interest, with each new algorithm bringing its unique twist on the challenges enumerated above. With these challenges in mind, aside from a few exemplary laboratories, reaction optimization algorithms have still not achieved widespread use in the field of basic chemistry.

To lower the barrier of entry between chemists and reaction optimization algorithms, user-friendly interfaces need to be designed between the chemist and the algorithm on multiple levels. EDBO+ provides a great example of this by providing a web interface as well as a relatively straightforward API. Our team has also designed Rxn Rover, a reaction automation platform that strives to create a user-friendly program to allow chemists to connect optimization algorithms directly to their reactors in a flexible way.

However, both approaches currently have issues. The EDBO+ web interface only works for the EDBO+ algorithm, which may not be the best choice for all reactions. Rxn Rover provides plugins for different optimization algorithms that are easily loaded in the program, but each of these plugins is difficult and time consuming to produce. On a level between these user interfaces and the optimization algorithms, there is a need for an abstracted interface that allows access to multiple optimization algorithms through common commands and formatting. It must also be relatively straightforward to add new optimization algorithm support to this interface.

Due to the myriad possible applications, implementations, data formats, and programming languages, it is likely impossible to define one abstract interface that will unilaterraly apply to all optimization algorithms in all fields. However, narrowing the scope to chemical reaction optimization, algorithms are commonly implemented in Python with one application in mind, optimize a reaction for x, y, z results. This significantly reduces the number of challenges in creating an abstract optimization algorithm interface.

CyRxnOpt aims to provide a single software interface that can provide access to many different optimization algorithms, mainly designed for use in chemical reaction optimization. It breaks chemical reaction optimization in to four main steps: Installation, Configuration, Training, Prediction. This allows users to program to a single software interface, simplifying the development of user-friendly tools and interfaces to lower the barrier of entry into reaction optimization.