Probabilistic Planning for Directed Evolution

Directed evolution (DE) has emerged as a powerful new methodology for the development of biocatalysts or proteins. The experimental protocols available to the practitioner of DE are expanding rapidly, and selecting the right combination of protocols that will lead to highly functional catalysts is currently an art, especially since the effects of experiments cannot be predicted reliably using current technology. Consequently, it is important to build appropriate models and optimization frameworks that use the information generated from the experiments to make selections amongst the protocols.

The purpose of the project is to implement a probabilistic approach based on value iteration, a simple probabilistic planning technique based on dynamic programming, and perform some experiments with it. An initial approach has already been worked out but still needs to get coded (in Matlab, Mathematica, or C, for example), tested, and refined. The deliverable is the code and a short writeup with some insights about the strengths and weaknesses of the approach and how it can be improved. This is an interdisciplinary project with researchers from chemical engineering and provides insights into the application of technology from artificial intelligence and operations research.