Advertisement

If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.

ENJOY UNLIMITED ACCES TO C&EN

Computational Chemistry

A protein-based neural network for cells

Scientists build artificial neural network to classify signals in cells

by Payal Dhar, special to C&EN
December 18, 2024

 

A vintage computer monitor with many dials and buttons shows a blue-and-white image of numerous cells. The background is made of two rectangles, in differing shades of blue, with light-colored threads running out from a central point on the bottom center of the image.
Credit: Madeline Monroe/C&EN/Shutterstock
In the protein-based neural network, the neurons are computational units.

Biological cells process data and perform computations all the time. They take inputs in the form of external stimuli and produce specific responses. Recently, scientists have been looking at ways to use that mechanism to program certain behaviors in cells. In one new study, researchers built an artificial neural network using proteins and used it to classify signals in cells. (Science 2024, DOI: 10.1126/science.add8468)

Generally, when neural networks perform a classification task, the system provides one of several outputs, depending on the data provided as inputs. In this case, the protein-based neural network allows the comparison of multiple inputs and provides a single output, such as programmed cell death or differentiation.

A neural network, whether biological or artificial, is a complex and layered system of interconnected nodes or neurons that work together to process the input, weigh options, and arrive at an output. The human brain is a biological neural network that receives stimuli and transmits signals to cells to keep the body working. In an artificial neural network, such as an application like ChatGPT, the neurons are computational units.

In this study, the neurons are proteases, or proteins that cleave other proteins according to certain signals. The inputs are de novo designed proteins. Without the inputs, the proteases are by default inactive, says Zibo Chen, a Westlake University biochemist and first author of the study. The proteases activate in the presence of their cognate protein partners.

Thus, the neural network’s activity is correlated with protease activity, Chen says. “Once the input—which are also proteins—come in, they activate the proteases.” The circuit takes into account weighted input summaries to decide which proteins to express, which in turn activates downstream neurons. Certain reversible binding interactions are also programmed in so the system can perform what is known as a winner-takes-all computation. Once this kind of process begins, there can be only one outcome.

“This is a very rudimentary neural network,” Chen says. “Our goal here is not to perform real neural network learning in our cells. . . . What we have is a carbon-based neural network, [which is] much easier to interface with actual biology.” Augmenting biological cells in this could eventually be used in cell-based therapeutics and diagnostics.

So far, computing via a biological neural network has been done only with DNA in test tubes or with gene circuits, Chen says. Computing with proteins is fairly new. “One thing we’re really excited about is its potential in cancer immunotherapy.” Because neural network architecture is also very scalable, the researchers hope to expand their circuit to sense multiple inputs at the same time.

“This work adds to the growing body of knowledge showing that complex computational approaches like neural networks can be implemented using synthetic biology and, in particular, proteins,” says North Carolina State University’s James Tuck, who was not involved in the study. “There are many potential downstream use cases of deploying such neural networks in living cells for both fundamental science and medical fields.”

Article:

This article has been sent to the following recipient:

0 /1 FREE ARTICLES LEFT THIS MONTH Remaining
Chemistry matters. Join us to get the news you need.