Overview
Our work of Dry Lab is mainly divided into three coherent parts: Explore, Design, and Validate.
We devised a new AI based approach called EGOAL to predict the gene expression from experimental condition, in order to explore the unknown regulation process of electricity production in Shewanella, and thus improve it with synthetic biology methods. In traditional AI models for science research, applications of data-oriented machine learning and knowledge-oriented machine reasoning are usually separated. However, in the case of human scientists, data-oriented perception and knowledge-oriented reasoning are both indispensable, and they work together to direct new discoveries. To merge them in a machine assistant of research, we employed abductive learning (ABL), a novel paradigm of knowledge representation learning proposed by our own supervisors. With ground knowledge of regulation and gene function, the regulation process can be predicted more precisely.
On the prediction of the regulation process, we then modeled the phosphorus metabolism network of Shewanella to explore the relationship between phosphorus metabolism and power production capacity. In detail, we built ODE1 and ODE2 (Ordinary Differential Equations) to illustrate the dynamics of mutiple components involved in gene expression, phosphorus metabolism, bioelectrogenesis, etc., through which having a better understanding of the underlying mechanism, and designing our Wet Lab experiment.
We alse built a phosphorus and energy cycles model to estimated how much phosphorus fertilizer this engineered bacteria could save us in space flight systems.
In order to put our ideas into practice, our hardware part designed an electric energy harvesting circuit for microbial fuel cells. This circuit is suitable for the collection of electrical energy from microbial fuel cells, which can collect the energy generated by microorganisms for subsequent use.
In addition, it is also our job to give positive feedback on the experimental data, suppot data mining and proper explanation of experiment results. We also utilize such experimental data to validate and tune our constructed models.