Compilation and Inference with Chemical Reaction Networks

Author: Poole, William

Year: 2022

Degree: Dissertation (Ph.D.)

Advisors: Winfree, Erik; Murray, Richard M.

Committee Members: Thomson, Matthew; Winfree, Erik; Murray, Richard M.; Phillips, Robert B.

Option: Computation and Neural Systems

DOI: 10.7907/x3qc-je74

Abstract

The successful advancement and deployment of technologies in the field of synthetic biology will require sophisticated computational infrastructure coupled with new theoretical ideas in order to more effectively engineer and reverse engineer biochemical networks. This thesis argues that the field of machine learning can inform the development of these underlying principles and techniques. First, software for compiling diverse chemical reaction network models of biological circuits from simple specifications is described. Second, three chemical reaction network implementations of a powerful machine learning model called a Boltzmann machine are analyzed and compared. Third, the class of detailed balanced chemical reaction networks are proven to be capable of probabilistic inference and, when coupled to a driven chemical system, autonomous learning. Finally, the use of machine learning to interpret and understand biological systems is explored in an experimental case study modeling E. coli cell extract metabolism.

Files