Machine learning is revolutionizing all areas of biology and industry but it is usually limited to a small number of users and situations. Tobias Erb, a team of researchers from the Max Planck Institute for Terrestrial Microbiology, has created METIS, a modular system for optimizing biological systems. Its versatility and usability are demonstrated by the research team using a variety biological example.
Although engineering biological systems is essential in biotechnology and synthetic biology has machine learning made it possible to use this technology in other areas of biology. But it is clear that algorithm improvement and application, which is a set of instructions that is computationally complex, is difficult to access. Not only are they limited by programming skills but often also insufficient experimentally-labeled data. There is an urgent need to find efficient ways to bridge the gap between computational and experimental work and machine learning algorithms that can be applied for biological systems.
Tobias Erb leads a Max Planck Institute for Terrestrial Microbiology team that has successfully dedemocratized machine learning. Their tool, METIS, was presented by the team and collaboration partners at the INRAe Institute of Paris in a recent publication in Nature Communications. The modular design of the application makes it easy to use on different biological systems with different equipment. METIS is short from Machine-learning guided Experimental Trials for Improvement of Systems and also named after the ancient goddess of wisdom and crafts Μῆτις, lit. “wise counsel.”
Less data required
Active learning, also known as optimal experimental design, uses machine learning algorithms to interactively suggest the next set of experiments after being trained on previous results, a valuable approach for wet-lab scientists, especially when working with a limited number of experimentally-labeled data. But one of the main bottlenecks is the experimentally-labeled data generated in the lab that are not always high enough to train machine learning models. “Active learning reduces the need to collect experimental data. However, we looked at various machine-learning algorithms. Amir Pandi, one the study’s lead authors, said, “We found a model which is even more dependent on data.”
The team demonstrated the versatility of METIS by using it for a range of applications including optimization of protein production and combinatorial engineered enzyme activity.2CETCH is the fixation metabolic cycle. The CETCH cycle was analyzed in a combinatorial area of 1025 conditions. They used only 1,000 conditions to find the most efficient CO.2The fixation cascade has been described.
Optimizing biological systems
The study’s application provides new tools that can be used to advance biotechnology, synthetic biology and genetic circuit design as well as metabolic engineering. Christoph Diehl is the co-lead writer of the study. “METIS allows researchers can either optimize their already synthesized or discovered biological systems,” he says. It can also be used as a combinatorial tool to understand complex interactions and hypothesis-driven optimizing. The best part is that it can be used to prototyping new-to nature systems.
METIS can be used as a modular tool that runs as Google Colab Python notebooks. It is available via a personal copy on a web browser. This work contains information that can be used to help users customize METIS for their own applications.