344
Productivity & Workflow355
Automation & Workflow224
Software Development251
Marketing & Growth192
AI Infrastructure & MLOps174
Writing & Content Creation203
Data & Analytics141
Photography & Imaging156
Design & Creative170
Customer Support131
Sales & Outreach125
Voice & Speech135
Education & Learning131
Operations & Admin87
A research team adapted AlphaFold to spot where CRISPR proteins make mistakes, then changed Cas proteins to cut off-target edits in tests.
In short: Researchers adapted Google’s AlphaFold to help redesign CRISPR gene-editing proteins, reducing unwanted “off-target” DNA edits in lab tests.
Gene editing is starting to show up in real medical treatments, but safety is a major concern. Even when scientists aim CRISPR at a specific spot in DNA, it can sometimes edit a similar looking spot elsewhere. These are called off-target effects, meaning the tool hits the wrong address.
A team of researchers based at several institutions in China reported a method to lower these mistakes. They used AlphaFold, a system that predicts how proteins fold and fit together (like guessing how a key is shaped so you can tell what locks it might open). Their idea was to compare how the Cas9 protein behaves when it binds to the correct DNA target versus a slightly mismatched one.
To do this, the researchers first built a large list of real off-target sites by running a CRISPR-based editor and then collecting DNA fragments that showed evidence of editing. They then used AlphaFold to model how Cas9, the guide RNA (a short “search string” that points CRISPR to a DNA sequence), and DNA sit together.
They focused on which parts of Cas9 physically touch the RNA and DNA. AlphaFold can estimate “contact probability,” meaning how likely two parts are close enough to touch. The team used this to flag amino acids (small building blocks of proteins) that shift during off-target binding, and called their setup “ContactSeek.”
They tested 23 small protein changes across 10 key positions. One redesigned Cas9 kept similar on-target activity, while off-target editing dropped from 28 percent to 5 percent in their tests. They also showed a similar approach could work with another CRISPR protein, Cas12.
If gene-editing therapies are going to be used widely, they need to be predictable. This work suggests AI models can help find specific spots to tweak so CRISPR is less likely to make rare but risky mistakes.
Source: Arstechnica