Cradle’s AI-powered protein programming platform levels up with $24M in new funding
Biotech and AI startup Cradle is finding success with its generative approach to protein design, landing big customers and a hefty $24 million of new investment.
The company exited stealth a little over a year ago, just as the hype around large language models was really heating up. Many AI companies in biotech train models to natively understand molecular structure; Cradle’s insight was that the long sequences of amino acids that make up the proteins in our bodies are akin to “like an alien programming language.”
It may not be possible for a person to learn that language, but an AI model could — and a person could work with that instead. While they still couldn’t just say “make a protein that does this,” they could ask which of 100 interesting proteins looks most likely to survive at room temperature or an acidic environment.
The approach seems to have caught the eye of major drug development companies like Johnson & Johnson and Novozymes. Creating a useful and functional protein from scratch is generally a pretty involved process, taking perhaps years and hundreds or thousands of wet-lab experiments.
Cradle says its tech can cut that time and the number of experiments required down significantly. Though it did not really substantiate claims of halving development time, it did provide an illustrative example from its in-house development.
They used their software to produce alternate versions of T7 RNA polymerase, an RNA production enzyme, that would be more resistant to high temperatures. Normally, they said, a team might expect under 5 percent of purposefully tweaked molecules to have the desired aspect, but 70 percent of the variants produced by Cradle showed increased stability. That’s the equivalent of running four or five such experimental runs in one.
The company exited stealth a little over a year ago, just as the hype around large language models was really heating up. Many AI companies in biotech train models to natively understand molecular structure; Cradle’s insight was that the long sequences of amino acids that make up the proteins in our bodies are akin to “like an alien programming language.”
It may not be possible for a person to learn that language, but an AI model could — and a person could work with that instead. While they still couldn’t just say “make a protein that does this,” they could ask which of 100 interesting proteins looks most likely to survive at room temperature or an acidic environment.
The approach seems to have caught the eye of major drug development companies like Johnson & Johnson and Novozymes. Creating a useful and functional protein from scratch is generally a pretty involved process, taking perhaps years and hundreds or thousands of wet-lab experiments.
Cradle says its tech can cut that time and the number of experiments required down significantly. Though it did not really substantiate claims of halving development time, it did provide an illustrative example from its in-house development.
They used their software to produce alternate versions of T7 RNA polymerase, an RNA production enzyme, that would be more resistant to high temperatures. Normally, they said, a team might expect under 5 percent of purposefully tweaked molecules to have the desired aspect, but 70 percent of the variants produced by Cradle showed increased stability. That’s the equivalent of running four or five such experimental runs in one.
Comments
Post a Comment