The baker’s yeast Saccharomyces cerevisiae has been helping humans turn sugars into alcohol and carbon dioxide for thousands of years. In the new generation of biotechnology, however, biologists are increasingly interested in oleaginous yeasts, among which Yarrowia lipolytica stands out for its ability to accumulate large amounts of oil-like compounds in its cells. An international team led by Estonian bioengineers set out to find a way to make the yeast produce valuable pigments, or carotenoids, in addition to lipids.

In its natural form, lipids make up between one-fifth and one-half of the yeast’s dry mass. By deliberately modifying its genes, the proportion of oil-like compounds can be increased to nearly 80%, while the cells can also be made to produce carotenoids that are not naturally present in the yeast. The researchers face a challenge, however, because thousands of metabolic processes work together as a network, and changing even a single gene can alter the entire system.

When researchers screen for the best genetic modifications, they rely on genome-scale metabolic models running on computers. These models describe the biological reactions taking place in a cell and help biologists test only the most promising variants in the laboratory. Each reaction in a cell is carried out by specialised proteins called enzymes, and the instructions for making these proteins are encoded in the yeast’s genes.

Previous models had gaps in the data describing enzyme properties. Biologists addressed this problem using the artificial-intelligence-based algorithm DLKcat. Based on existing information, the computational tool can predict the properties of almost all Yarrowia lipolytica enzymes.

The study’s lead author, Tallinn University of Technology researcher Juliano Sabedotti de Biaggi, explained that artificial intelligence accelerated enzyme analysis several hundredfold. “With previous mechanistic models, calculating the properties of a single enzyme took several days, whereas DLKcat was able to provide the same information for hundreds of enzymes in less than an hour,” the researcher said. This made it possible to study and compare enzymes on a much larger scale and carry out analyses that had previously been computationally impractical.

In laboratory experiments, the researchers also collected extensive proteomics data and fed these data directly into the new computational model. The data reflect which proteins the yeast produces at different stages of its life cycle and in what quantities. This resulted in a new and, according to the article, more detailed metabolic model, eciYali5-GEM.

The needs of a yeast cell change dramatically over the course of its growth. A gene that stimulates lipid synthesis during an early growth phase may have no effect later on. The new model can therefore help researchers identify the right time to increase or decrease the activity of a particular gene in the cell.

When the researchers compared the computational model’s predictions with laboratory experiments, they found that the new approach was able to reproduce previous experimental results. The algorithm not only confirmed earlier findings but also identified completely new metabolic engineering targets. This gives researchers a clearer indication of which genes to target first.

Carotenoids and lipids are produced in yeast through related intracellular pathways. When researchers direct the cell’s resources towards one pathway, fewer resources are inevitably available for the other. The detailed computational model can help identify a balance at which the yeast is able to produce both valuable compounds at the same time.

Artificial intelligence and mathematics do not solve all of biologists’ problems, however. The researchers found that some of the genes recommended for modification by the algorithm did not produce the expected results in laboratory experiments. The actual metabolism of the yeast remained too complex for current computational models to fully capture.

“This shows that the metabolism of the yeast cell is more complex than current models are able to describe. However, even unsuccessful predictions can help improve future models,” Sabedotti de Biaggi added.

The researchers see broad potential for the new approach across industrial biotechnology. Smart computational models can help biologists design cells capable of consuming industrial waste streams and converting them into useful compounds. The authors wrote in their research article that similar computational approaches could eventually be adopted by researchers studying other microorganisms.

Efficient cell factories could help humanity reduce its dependence on petroleum and intensive agriculture. Artificial intelligence could help biologists put yeasts to productive use, allowing sustainable biofuels, novel cosmetics and food supplements to move from laboratories to factories sooner. The combined power of biology and mathematics could bring changes in everyday consumption one step closer.

The study was published in the journal Applied Microbiology and Biotechnology.