AI model could help create synthetic blood plasma • The Register

Scientists are using AI algorithms to design new materials, including synthetic proteins to make fake blood plasma and biological liquids found inside of cells.

These artificial substances were able to support natural proteins and boost their properties, say researchers. The AI-made blood plasma, for example, could dissolve and keep real protein biomarkers at room temperatures without the need for refrigeration.

In another experiment, a solution containing a mixture of fake polymers to mimic cytosol, the liquid found inside cells, was produced inside test tubes. When ribosomes, molecules containing RNA, were added, they continued to function normally producing proteins as if they were operating inside a real cell.

Ting Xu, professor of chemistry as well as materials science and engineering at the University of Berkeley, led the research and believes AI can design new polymers to augment biological proteins. Proteins are a type of polymer, molecules made up of a sequence of smaller, repeating units. In proteins, the building blocks are 20 different amino acids.

"Basically, all the data shows that we can use this design framework, this philosophy, to generate polymers to a point that the biological system would not be able to recognize if it is a polymer or if it is a protein," Xu said in a statement. "We basically fool the biology.

The deep learning model, a modified variational autoencoder (VAE), learned to encode features mapping protein structures to their function, and was used to design new materials with specific properties. "We look at the sequence space that nature has already designed, we analyze it, we make the polymer match to what nature already evolved, and they work," Xu said.

"It was trained using natural proteins by dissecting proteins into short segments. By pooling the chemical characteristics of these segments that have been selected evolutionarily, we develop a blueprint for the polymer to mimic," she explained to The Register.

The system, described in a paper published in Nature, allows researchers to design hybrid biological systems or imbue existing materials with new properties. Scientists could make new plastic polymers that naturally degrade over time to combat waste, for example, or develop synthetic proteins to deliver drugs or enhance biological processes like photosynthesis.

Xu believes scientists need to start designing synthetic proteins in new ways, and place less emphasis on trying to mimic natural structures. Unlike real proteins, the AI-made polymers were made up of just five or six different amino acids, making them less complex and easier to synthesize.

"The Food and Drug Administration hasn't approved any new material for polymer biomaterials for decades, and I think the reason is that a lot of synthetic polymers are not really working — we are pursuing the wrong direction," she said. "We are not letting the biology tell us how the material should be designed. We are looking at individual pathways, individual factors, and not looking at it holistically." ®