Creating a new drug was once a gruelling exercise in trial-and-error lab experiments. Artificial intelligence is now accelerating the discovery process, helping human researchers to simulate tests, creating novel designs for new medications, and improving the quality of clinical data.

“AI is revolutionizing drug discovery and is significantly cutting the time-to-market for a new medication,” says Felice Verduyn-van Weegen, a partner in EQT’s Life Sciences team.

Life sciences companies are enhancing their operations by integrating AI throughout their operations. HotSpot Therapeutics, to give one example, is using an AI-powered drug discovery system to find “hotspots” in proteins that are critical to a cell’s function and could effectively be reprogrammed by drugs. CluePoints is using advanced statistical modelling to improve the quality of clinical trial data.

Egle Therapeutics is using machine learning, a form of AI that generates predictions from large datasets, to power its research into white blood cells known as regulatory T cells, or Tregs. These cells deliberately suppress the body’s immune response to stop it overcompensating and causing damage – but some cancers trick particular Tregs into entering immune suppression mode, so Egle is developing drugs that disable these malfunctioning cells.

Egle’s engineers program machine learning algorithms to process huge databases of sequenced tumors. They tell the algorithms to extract Treg data from the tumor data and ask it to determine why these Tregs support the cancer while other Tregs do not. The software then produces a list of around 200 ‘target’ molecules that are only present on the faulty Tregs, allowing researchers to design drugs that attach to the target molecules and destroy or impair the cells.

“We wouldn’t be able to do this type of work in the lab,” Fiorella Kotsias, Chief Scientific Officer at Egle, says. “We’re talking about millions of cells per tumor, and around 200,000 genes per cell. That gives you an idea of the size of the datasets we’re using.

“Machine learning allows us to make sense of and extract ideas from a massive amount of data. But we still need human researchers to validate these ideas by working on real-life tumor samples in a lab. Our work in the lab involves all the classic experiments you might expect, but we also ensure that each of our experiments is sequenced, and we use that data to enhance our research.”

Putting the AI in antibodies

Another area where AI has proved to be particularly powerful is antibody design. Researchers are using AI-powered software simulations to design new antibodies that can attack four or more targets – which is difficult to do efficiently with traditional lab-based testing. The software allows researchers to simultaneously tweak a wide range of parameters for each antibody candidate and run simulations to see whether the candidate succeeds or fails. This allows researchers to narrow down the list of candidates and focus on the most promising ones in the lab.

EQT was an early mover in AI investing, meaning that its portfolio companies outside life sciences are positioned to help those within the sector – and vice versa. “AI tech and AI good practices aren’t sector-specific,” Verduyn-van Weegen says. “We want to ensure that best-in-class technology is being used by all our portfolio companies to ensure their speed and execution is at the next level.”

She says EQT is cautious about investing in life sciences companies that describe themselves as pure-play AI. “We haven’t seen a huge amount of success so far from these sorts of companies in terms of discovery, exits to public markets, and mergers and acquisitions. We prefer what you might call ‘traditional’ life sciences companies that are generating good data and are already leveraging AI to improve their operations or have the potential to do so with our support.”

Data needs

AI has its limitations, of course. Most contemporary AI – including OpenAI’s ChatGPT – draws heavily from public data, which is not wholly reliable. This means that outcomes from these models need to be checked very carefully by humans. There is also a specific concern about scientific data: negative results from clinical trials are not often published. This means that AI used in science can skew towards the positive because it has no perspective on negative findings.

In drug discovery, the AI of the future could simply be asked to find the best target for a particular disease and design a drug that effectively interacts with it, with little or no lab testing. However, we’ll need much more experimental data to feed the algorithms to achieve this. Assuming enough data can be created, we may be able to reach a world of fully personalized medicine, whereby AI designs bespoke drugs for individual patients. With this sort of power, AI could become less of a research tool and more of a foundational business model.