Tell this AI what you want a protein to do—and it designs one

New artificial intelligence (AI) technology called Pinal has been developed by researchers at Westlake University in China. This AI model can generate candidate proteins from natural language prompts, eliminating the need for users to have extensive knowledge of molecular structures and geometric constraints.

Pinal uses two main modules to operate: T2struct and SaProt-T. The first module, T2struct, takes the natural language description of the protein's desired function and builds a 1D structural blueprint. The second module, SaProt-T, reads this blueprint and generates a matching amino acid sequence based on the original text prompt.

Pinal's standout feature is its ability to create multiple structural blueprints, design the best sequence for each, and score each pair to determine the optimal final design. Training Pinal required an extensive dataset of 135 million protein sequences and 1.7 billion protein-text pairs. The researchers tested Pinal by asking it to design four different proteins - a fluorescent protein, a plastic-breaking enzyme, an alcohol dehydrogenase, and a metabolic H-protein.

For the fluorescent protein, Pinal generated 50,000 sequences, which were filtered down to 29 for testing. Only one of these sequences fluoresced under illumination. While Pinal has shown promise in producing functional metabolic H-proteins, precise control of protein properties remains a challenge. For example, creating a fluorescent protein with a specific emission wavelength is still difficult.

However, Pinal excelled in producing a metabolic H-protein that boosted the output of a key carbon-capturing process by 1.7 times that achieved using the natural E. coli protein.