A robot scientist for yeast systems biology

Lead Research Organisation: Aberystwyth University
Department Name: Computer Science

Abstract

The application of Artificial Intelligence to scientific research is growing in importance because of the increasing power of computers increased use of laboratory automation. The need for automation is particularly important in the branch of science known as systems biology, where scientists are trying to understand how genes work together to form living cells. Applying Artificial Intelligence to laboratory robotics we have previously developed a Robot Scientist that generates hypotheses about the function of particular genes in bakers yeast, and then designs and carries out experiments to test them. Baker's yeast (Saccharomyces cerevisiae) is used as a model organism for human cells, as it easier to grow and experiment with, and relatively simple (approximately 6,000 genes compared to approximately 30,000 in humans). We set the robot the problem of discovering the function of different genes in yeast. The functions of about 30 per cent of the genes in yeast are still unknown and, with many of these genes thought to be common to the human genome, they could prove to be medically important in the future. The research involved using knockout strains of yeast that have had one gene removed. By observing how the yeast grows, or doesn't grow, on defined chemical substrates, it is possible to start establishing different possible functions for the gene being investigated. It is like trying to understand what the different components in a car do by removing them one by one. The robot scientist generates a set of hypotheses from what it knows about yeast metabolism and then plans an experiment that will eliminate as many hypotheses as possible, as fast and as cheaply as possible. It conducts experiments by dispensing and mixing liquids and then measuring the growth of yeast using an adjacent plate reader that feeds the results back into the system. The robot then evaluates the results against the set of hypotheses, generates new hypotheses, and the process starts again - the same type of cycle human scientists use to understand the world. In this proposal we plan to extend the Robot Scientist in a number of ways: We shall use the Robot Scientist to develop a model of the whole of yeast metabolism. We shall develop new reasoning mechanisms for the robot. We shall use a much bigger robot capable to running many more experiments automatically.

Technical Summary

We have developed the first physically implemented active learning system aimed at scientific discovery - the Robot Scientist. This system can automatically: originate hypotheses to explain data, devise experiments to test these hypotheses, physically run the experiments using a laboratory robot, interpret the results, and then repeat the cycle. We believe that the Robot Scientist is currently the world's most advanced system for automating scientific experiments. The completion of the sequencing of the key model genomes, and the rise of post-genomic technologies: microarrays, proteomics, metabolomics, etc. are producing floods of data replete with undiscovered knowledge. This data has opened up the prospect of modelling cells (mathematically/computationally) in silico in unprecedented detail. Such models are essential to integrate our growing biological knowledge and have the potential to transform medicine and biotechnology. To build systems biology models, will of necessity, require the execution and analysis of unprecedented numbers of wet experiments; as models can ultimately only be confirmed or rejected by comparison with real biological systems. The execution and analysis of this vast number of experiments will, in turn, demand greatly increased automation. Therefore, we argue that systems biology will increasingly require a Robot Scientist like approach. We therefore propose to apply the Robot Scientist methodology to the challenging domain of systems biology. Specifically, we will use the Robot Scientist to develop and refine a whole genome model of yeast (S. cerevisiae) metabolism and growth. Yeast was the first eukaryote sequenced, and has arguably the most known of any organism about its transcriptome, proteome, and metabolome. Therefore, yeast is the ideal candidate organism for systems biology. For this we will utilize a new 450,000 pound bespoke robotic hardware system that we have designed that can run approximately 3,000 simultaneous quantitative growth experiments a day. To utilize this hardware to allow to automatic construct an experimentally verified logical model of yeast metabolism the Robot Scientist requires: its scientific inference mechanisms to be extended to allow it to more intelligently infer new hypotheses and the experiments to test them; these mechanism to be integrated with changes in the system model and background knowledge of yeast bioinformatics; and software to be developed to control the new hardware. We will also investigate the feasibility of extending the Robot Scientist's logical model into a hybrid quantitative one, and thereby enable the Robot Scientist to make quantitative inferences.

Publications

10 25 50
publication icon
Coutant A (2019) Closed-loop cycles of experiment design, execution, and learning accelerate systems biology model development in yeast. in Proceedings of the National Academy of Sciences of the United States of America