New AI tools let researchers design proteins and DNA using plain text

In short: Researchers introduced Pinal, an AI system that can design new proteins from plain-language prompts, and Hugging Face is hosting a similar tool for DNA called Carbon X.

What happened

A research team based in China developed a tool called Pinal that takes natural language prompts, like a short written request, and returns a proposed protein sequence. Proteins are the tiny machines inside living things, and their sequence is like a recipe that tells a cell how to build them.

Most protein design software today needs detailed expert input, like the exact 3D placement of key building blocks. Pinal tries to skip some of that by learning from large databases of proteins that already have labels describing what they do.

In tests, Pinal could produce working proteins, but the hit rate was low. For example, when asked to design proteins that glow like the green fluorescent protein (GFP), it generated 50,000 candidates, which were filtered down to 29 lab-made proteins, and only one actually glowed.

Separately, Hugging Face and partners released Carbon X, a model that works with both English and DNA sequences. DNA is the long string of letters in cells, like an instruction manual. Carbon X can take a DNA snippet and suggest what might come next, or answer questions like what could happen if one DNA letter is changed.

Why it matters

These tools may make early-stage biology research easier to start, since people can ask for a function in plain language instead of learning complex software inputs. But they do not remove the need for lab testing, because many generated sequences will fail, and checking them still takes specialized equipment and expertise.

Source: Arstechnica