A UK startup aiming to solve key challenges in life sciences through AI has unveiled the “world’s first AI models for programmable gene insertion”, as it confirms Nvidia has invested in the startup. London-based Basecamp Research harnesses biodiversity to design new medicines.
It does this by building its own AI models and today it has unveiled a family of AI models capable of programmable gene insertion, which it says offers a new way to replace faulty genes and reprogramme cells for therapeutic use. It is hoped the models can propel a new generation of treatments for cancer and other inherited diseases.
John Finn, chief scientific officer, Basecamp Research, said: “We believe we are at the start of a major expansion of what’s possible for patients with cancer and genetic diseases. By using AI to design the therapeutic enzyme, we hope to accelerate the development of cures for thousands of untreatable diseases, potentially transforming millions of lives.”
The models have been released with a paper, co-authored by Nvidia, Microsoft and academia, and are being unveiled at Nvidia's presentation at the JP Morgan Healthcare Conference in the US. The new AI models cost tens of millions of pounds to produce, according to Basecamp Research CEO and co-founder Glen Gowers, and have been built in collaboration with Nvidia, whose GPUs the models were trained on.
Gowers said: “It is very similar to the GPT style of models. In fact, the scale of this model is roughly parallel to GPT4 in terms of the training data it has seen.” The models are trained on Basecamp Research's proprietary genomics dataset, which scientists have gathered from around the world.
According to the startup, the models learn the language of DNA and patterns of evolution, allowing the algorithms to design new, programmable therapies for cancer and genetic disease. Gowers said: “In this model, what we test it on is multiple different disease types and multiple different types of medicine.”
He added: “We can type in a pathogen that we want to kill and with about 97 to 100 per cent accuracy, it then produces a peptide that, currently kills in the lab, that pathogen.”
On being able to build models on a much smaller budget compared to US AI labs, Gowers said: “You want to really carefully pair the amount of data with the right size of model. It’s not just the case that you throw larger and larger amounts of compute at it.
“We ourselves built the training dataset and we trained the model, which means we can very, very tightly make sure that we have the optimal combination of both.” On the post-training of the models, it used lab-generated data, which helped keep costs down, Gowers said.
The models are not being released as a chatbot, but will be used by the startup in partnership with pharmaceutical companies to design therapeutics. On applications, Gowers pointed to its advantages over CRISPR-based approaches.
He said: “What we are showing in this paper is the ability to do large edits, large insertions, which opens the door to tens of thousands of genetic diseases that today wouldn’t be treatable by CRISPR.”
Basecamp Research has also confirmed that it has secured investment from Nvidia’s VC arm NVentures, in a pre-Series C round, but did not declare how much the chip giant has invested.
Basecamp Research was one of the UK startups named last year by Nvidia CEO Jensen Huang, which he said Nvidia planned to invest in, along with others such as Revolut and Wayve, as part of a £2bn investment to support the UK AI industry.
Basecamp Research has raised around $85m to date.
Its investors include VC firms Singular, True Ventures and Hummingbird Ventures and Paul Polman, the former Unilever CEO.
Would you like to write the first comment?
Login to post comments