Medically, AI is helping us with everything from identifying abnormal heart rhythms before they happen to spotting skin cancer. But do we really need it to get involved with our genome? Protein-design company Profluent believes we do.
Founded in 2022 in Berkeley, California, Profluent has been exploring ways to use AI to study and generate new proteins that aren't found in nature. This week, the team trumpeted a major success with the release of an AI-derived protein termed OpenCRISPR-1.
The protein is meant to work in the CRISPR gene-editing system, a process in which a protein cuts open a piece of DNA and repairs or replaces a gene. CRISPR has been actively in use for about 15 years, with its creators bagging the Nobel prize in chemistry in 2020. It has shown promise as a biomedical tool that can do everything from restoring vision to combating rare diseases; as an agricultural tool that can improve the vitamin D content of tomatoes, and slash the flowering time of trees from decades to months; and much more.
Over the years, CRISPR has relied on proteins found in nature to get its DNA-slicing work done. In particular, it employs the Cas9 protein, which was originally derived from bacteria, to open strands of DNA in order to alter gene sequences.
Thinking that AI could improve on the CRISPR system in much the same way it is beginning to improve in multiple arenas, the Profluent team created a database of 5.1 million Cas9-like proteins which, the company says, bested any previous such database by 2.7 times overall and 4.1 times for Cas9 proteins specifically. An AI large language model was then trained on the database and asked to create potential proteins that could be used instead of Cas9 in CRISPR work.
After Profluent's team guided the system in whittling down the results from 4 million sequences, it eventually landed on OpenCRISPR-1.
The new protein performed as well as Cas9 in tests, but with one noticeable difference: it reduced the impact on off-target sites by 95%. This means it was much more precise, working pretty much only where needed and not causing any residual damage to the chain of DNA.
In an unusual move, Profluent made its protein available to the scientific community (thus the "open" part of OpenCRISPR-1).
“Today’s announcement is a watershed moment and the beginning of what we hope will be an iterative process as we embark on this next generation of building genetic medicines,” said Peter Cameron, Vice President and Head of Gene Editing at Profluent. “We encourage the gene editing community to pressure test OpenCRISPR-1. If there are particular features that could be improved for a specific application, we’d like to know and can collaborate to optimize those properties.”
Cause for alarm?
On its surface, using AI to alter DNA in any way sets off alarm bells. But in this case, thoughts of "insect children" or "Dr. Frankenstein" as raised on CEO Ali Madani's X post about the tech, are largely overblown. Profluent simply used AI to do what it does best, digest enormous amounts of data and reach conclusions faster than any human being could hope to do.
Rather than dubious genetic experiments, the company feels that the new technology will lead to rapid and accurate genetic treatments for common diseases.
"We hereby publicly release OpenCRISPR-1, a highly performant AI-generated gene editor, to facilitate broad, ethical usage across research and commercial applications," said Profluent team members in a blog post. "We aim to advance innovation and development in the gene editing community, to bring new treatments to patients with major unmet needs."
Profluent's breakthrough has been released on the preprint server bioRxiv.
Source: Profluent
Jk. Amazing job with making it OPEN source! I can’t wait to edit it and release my version.