Machine learning for drug design, molecular machines and evolvable artificial cells

Abstract

An artificial cell is a complex chemical system with many components fabricated and assembled in the laboratory. The molecular components can be designed to interlock in a variety of different way to achieve the emergence of minimal life [1][2]. One experimental design is composed of three modules or sub-systems: lipid vesicles, a metabolic system and a cell free expression system. Due to the high number of molecular species and their non-trivial interactions in an artificial cell any prediction of the emerging properties in this high dimensional space of compositions is extremely difficult. Previously we have developed and used a machine learning process Evo-DoE (Evolutionary Design of Experiments) coupled with a robotic workstation for liquid handling to optimize a liposomal drug formulation [3] as well as a cell free expression system for the synthesis of the GFP (green fluorescent protein in vitro) [4]. In addition we have results of vesicle fusion providing a protocol to design a life-cycle for evolvable artificial cells. Now we propose how our technologies could be used to optimize artificial cells.