Anthropic's Claude AI designs new proteins for drug discovery
What's the story
Anthropic, a leading American artificial intelligence (AI) company, has successfully leveraged its Claude AI model to design proteins for potential drug discovery. The firm recently published a blog post detailing the experiment where Claude was tasked with creating protein "binders" from scratch for 15 different targets. The results were promising, with the AI successfully designing binders for 14 of them.
Teamwork
Collaboration with biotech firms
The successful protein designs were a result of a collaboration between Anthropic, Adaptyv Bio, and Twist Bioscience. The two companies independently built and tested the proteins designed by Claude.
In total, the researchers produced 1,320 designs across multiple experiments. Out of these, lab testing identified 354 binders against 14 of the 15 targets.
Some successful designs even showed particularly strong results.
Potential
Exceeding expectations
The experiment with Claude produced high-affinity binders against at least six targets.
For at least four of these targets, some of Claude's binders matched or exceeded the best previously reported binding strength.
This is a significant achievement, as stronger binding can be crucial in drug development, potentially allowing treatments to work at lower doses.
Performance
Hit rates
Anthropic reported an overall hit rate of 35.1% when its Mythos Preview model worked on individual targets in separate sessions.
A hit rate is the percentage of designs that actually bind to the intended target.
When working against all targets simultaneously during a 48-hour session, Mythos Preview had a hit rate of 26.7%, while Opus 4.8 reached 22.6%.
These rates are significantly higher than the typical 10%-15% seen in today's protein design campaigns.
Caution
Early research results
Despite the promising results, Anthropic has cautioned that Claude's protein designs should be viewed as an early research result.
The company emphasized that these findings do not prove AI can independently create new medicines.
This is an important distinction to make, as the field of AI-assisted drug discovery continues to evolve and develop.