Chemical computer can recognize patterns and multitask

A chemical computer made up of a network of enzymes can perform various tasks, such as measuring temperatures or identifying substances, without needing to be rebuilt each time. This makes it more of an adaptive biological system than a digital circuit, and offers the promise of linking computers to biology.

Living organisms contain molecular networks that constantly integrate chemical and physical signals, for example when cells sense nutrients, hormones or temperature changes and adapt to stay alive. For decades, researchers have tried to imitate this in various ways, such as building logic gates from DNA, but most of these artificial systems were too simple, too rigid, or too difficult to scale.

Now, Wilhelm Huck of Radboud University in the Netherlands and his colleagues have taken a different approach. Instead of programming each chemical step, they built a system in which enzymes interact freely, creating complex behaviors that allow them to learn to recognize patterns of chemical inputs.

The team’s computer uses seven different types of enzymes loaded onto tiny hydrogel beads packed into a small tube. A liquid flows through this tube and can be injected with short chains of amino acids called peptides, which serve as “input” to the computer. As peptides pass through enzymes, each enzyme naturally attempts to cut them at specific sites along the peptide chain. But once an enzyme makes a cut, the shape of the peptide and the available cutting sites change, which can open or block opportunities for other enzymes.

Because one reaction can fuel the next, enzymes create an ever-changing chemical network, producing distinctive patterns that the system can interpret. “We can think of enzymes as… hardware and peptides as software. [that] allows you to solve new problems based on inputs,” says Dongyang Li of the California Institute of Technology, who was not involved in the work.

For example, temperature affects the speed at which each enzyme works; at higher temperatures, some enzymes accelerate more than others, thereby changing the mix of peptide fragments in the system’s production. By analyzing these peptide fragments using a machine learning algorithm, researchers could link these fragment patterns to specific temperatures.

Because different chemical reactions occur on different time scales, the system naturally retains a sort of “memory” of past signals, allowing it to recognize patterns that unfold over time. For example, it could differentiate between fast and slow light pulses, meaning it not only reacts to inputs, but also tracks their change.

The result is not a static chemical circuit, but rather a dynamic, multitasking chemical computer that processes signals like a living system. “The same network handled multiple tasks—chemical classification, temperature detection with an average error of about 1.3°C between 25°C and 55°C, pH classification, and even responding to the periodicity of light pulses—without rethinking chemistry,” Li says.

Researchers were surprised by the computer’s performance, given its small size, and Huck hopes that a more advanced system could one day be used to translate optical or electrical signals directly into chemical signals, allowing it to respond in the same way as living cells. “We only used six or seven enzymes and six peptides,” he explains. “Imagine what you can do with a hundred enzymes.”