Anthropic’s new biology tools aim to expedite the R&D process

A little over a year ago, Anthropic CEO Dario Amodei published an essay on what he saw as some best-case scenarios for AI. Much of it focused on potential scientific progress in biology, medicine, and neuroscience.

Now the AI company is making its first foray into that space with a new tool called Claude for Life Sciences, which upgrades the model to support biotech R&D labs with tasks like literature reviews, lab protocols, data analysis, and regulatory compliance.

The company is hardly alone in offering AI tools that specifically cater to scientists. Google rolled out a tool called Co-Scientist earlier this year; OpenAI is currently building a new team focused on AI for scientific discovery; and startup Lila Sciences has raised more than half a billion dollars as of last month to automate the scientific process.

Lab assistant: Eric Kauderer-Abrams, Anthropic’s head of life sciences, told us that what sets Anthropic’s tool apart is its compatibility with existing lab tools and its features across the entire scientific process.

“A lot of other players in the AI and bio space are focusing more on the early stage, you know, isolated research problems…In our experience, it’s only part of a much larger picture of developing something and bringing it to market and impacting patients,” he said.

While the tool doesn’t involve any new models, Anthropic has connected Claude with common biotech platforms like PubMed, Benchling, BioRender, and Synapse.org. “What this does is give Claude context to be able to participate with you throughout the whole process,” Kauderer-Abrams said, adding that the goal is to eventually task Claude with running experiments on its own.

“Once Claude is superhuman at all of these tasks around running the actual experiment, then it’s time to close the loop and make sure that we can actually have Claude assist, and in some cases, take on the execution of running these experiments in the lab as well,” he said. “And that’s a bit further out.”

Progress so far: Since Google DeepMind’s AlphaFold2 protein prediction breakthrough in 2020, which won the Nobel Prize last year, hype has grown around AI’s potential to accelerate drug discovery. Is that progress playing out as it was expected to a few years ago?

Kauderer-Abrams emphasized the long timeline that it takes to bring a drug from R&D to the market. “It’s too early to really discern one way or another…We have the first wave of AI-designed drugs entering early clinical trials now. And so with the timelines of drug development, it’s too early to call it,” he said.

Anthropic’s new tool aims to help researchers move down the pipeline from that initial drug discovery phase, Kauderer-Abrams said.

“Even if these new tools are incredibly successful at designing the new drugs, it doesn’t address the remaining, you know, 90 steps required to bring a new drug into the market,” he said. “And so I think what we’re going through now is the rest of the process.”

So where should we look for the first impacts of AI drug discovery in the healthcare system?

“There are certain rare diseases where, by construction, the development process can be shorter just due to the risk-benefit analysis for these clinical trials,” Kauderer-Abrams said. “I think it’ll be really interesting to watch how this first batch of these AI-first biotech startups do.”