Self-driving autonomous machines to engineer novel enzymes

Can one design and automate a computational and experimental platform such that each platform iteratively guides and drives the other to achieve a pre-determined goal? Rapp and colleagues (2024) describe just this possibility in a paper that details a prototype of a self-driven laboratory that can navigate autonomously to yield an engineered enzyme with a desired attribute. This laboratory, rather, the automated protocol, is referred to by an acronym - SAMPLE. This refers to Self-driving Autonomous Machines for Protein Landscape Exploration. The paper describes a prototype involving the engineering of a glycoside hydrolase for enhanced thermostability. The 'brain', the computational component behind this automated system, was designed to learn protein sequence- function relationships from a curated dataset. These designer proteins were then evaluated by a fully automated robotic system that could synthesize and experimentally characterize the designed protein and provide feedback to the agent, i.e., the computational component, to fine-tune its understanding of the system. The SAMPLE agents were thus designed to continually refine their understanding of the protein landscape by actively acquiring information in the search process. As this intelligent agent learns protein sequence-function relationships from a curated, diverse dataset, this feedback is crucial to refine landscape exploration and the design of new proteins based on the updated hypothesis. In this prototype, four SAMPLE agents were tasked with this goal. The goal of each of these agents was to navigate the glycoside hydrolase landscape and identify enzymes with enhanced thermal tolerance. Differences in the search behavior of individual agents primarily arise from experimental measurement noise. However, despite differences in their search behavior, all four agents could converge on a thermostable glycoside hydrolase - a remarkable feat as it apparently did not need any human intervention.