Scientists have developed an AI system that can produce synthetic enzymes from scratch.

The model was created by simply loading the amino acid sequences of 280 million different proteins into the model. The model was then given a few weeks to process the data. The model was then modified by feeding it 56,000 sequences from five distinct lysozyme families and some background information on these specific proteins.

The model swiftly produced a million sequences, out of which the study team chose 100 to test based on how closely they matched the sequences of genuine proteins and how realistically the underlying amino acid "grammar" and "semantics" of the AI proteins were.

The scientists created five artificial proteins to test in cells and compared their function to an enzyme called hen egg white lysozyme, found in the hen egg white lysozyme (HEWL), from this initial batch of 100 proteins that Tierra Biosciences evaluated in vitro. Human tears, saliva, and milk all contain similar lysozymes that operate as antibacterial and antifungal defenses.

Despite having only about 18% sequence identity, two synthetic enzymes were capable of degrading bacterial cell walls with activity equivalent to HEWL. About 90% and 70% of the known protein may be found in each sequence.

The team discovered that the AI-generated enzymes displayed activity even when as little as 31.4% of their sequence resembled any known natural protein. A natural protein can become inactive with just one mutation, but in a different round of screening, the team discovered that this was not the case with the AI-generated enzymes.

The AI could even determine how the enzymes should be structured by simply studying the raw sequence data. X-ray crystallography measurements revealed that the manufactured proteins atomic structures were correct, despite their sequences being novel. 

(Source: https://www.sciencedaily.com/releases/2023/01/230126124330.htm)