Perspectives: AI can revolutionize drug discovery, but we need software that’s up to the job

By: Julien Hobeika, Naza Metghalchi and Pierrick Cudonnec

Through our “Perspectives” series, we aim to deep dive and pose challenging questions that spark curiosity and stimulate debate.

In this piece, we explore the potential for AI to transform drug discovery and emphasize the importance of having software that can meet this challenge.

Over the last decade, the life science industry has witnessed an explosion of multimodal data

• The rise of clinical medical records in hospitals and new imaging technological video capture has been a huge factor in this explosion
• Increased and improved omics data and human genome sequencing has given us a deeper understanding of the biological system
• We’ve also seen a huge proliferation in the number of bio-banks and available data sets
• Today’s “publish or perish’’ academic environment — every minute, two research papers are published on PubMed — has led to exponential growth in the number of research papers available
• The pervasiveness of smartphones and other devices has provided us with vast data which can be used in clinical development

You would think that all this data would become more actionable at scale, with AI infiltrating drug discovery and development — the business that underpins the entire life sciences industry.

This has not been the case. In fact, over the last decade, the industry has struggled to discover, handle and produce new drugs. Clinical trial success rates have plateaued at below 10%, while the cost of bringing a drug to market has risen steeply. Two decades ago the average cost was $800m. Today it’s $2.3bn.

Despite this, many in the industry are still banking on AI to transform drug discovery. Do they need a reality check?

Where we are now: Big Pharma is still figuring out AI

Big pharma knows the potential of AI, and is looking to hire more AI scientists. However, the sector doesn’t have the same appeal to talent as native AI or biotech companies. Until it does, decision makers in the sector will need to leverage tooling and software so that less technical people can access and utilize these ever-expanding datasets. We’re already seeing the emergence of companies like Seqera to simplify the complex processing of omics data.

This lack of native talent also means the industry’s main exposure to new technologies comes via external partnerships with companies like Benevolent AI — an expensive approach, but nothing Big Pharma’s deep pockets can’t handle. Down the line, if that technology becomes fundamental, the companies can look to acquire it or build it themselves.

All of this is happening amongst board-level pressure to be more generative AI and data-driven, especially at the research and discovery stage. ‘In the last four years, boards are asking what data has been used to support a target for future development,’ one top 20 Big Pharma firm board member told us.

We won’t have to wait much longer to see the initial results as the first wave of therapeutic candidates designed and optimized by AI/ML tools are currently making their way through clinical trials.

The question now is whether we will see a new generation of companies leveraging this climate to disrupt the pharma industry, or is there room to build and scale software that enables existing pharma companies to 5x their success rate?

The software venture case

It’s the latter question that we find most interesting.

To find out, we have been meeting with dozens of companies building at the frontier of AI and drug discovery. From these meetings, it became clear that companies were taking one of two approaches: selling results and discoveries (bio-tech/tech-bio) or delivering SaaS products for a narrow section of the drug discovery funnel.

Both face, in our view, hard challenges. Let us explain.

Approach one: Selling results

The first approach sees companies focusing on leveraging an underlying technology to sell “results” to the pharma companies in the form of “insights”, “simulations”, “validated targets” or “compounds”. As this aligns with big pharma’s current preference for outsourcing research to edge the risk, this usually grows into revenue pretty quickly. Broken down, that often means a few million dollars upfront, plus a huge payout later in the form of royalties or fixed success fees. However, the challenge here is that the success of the “IP” is not necessarily up to the startup. Sometimes, pharmaceutical companies make decisions based on portfolio prioritization and end up killing successful programs which could have yielded a large upside.

For VCs looking to invest in these startups, genuine sector knowledge and expertise is crucial. There are many specific factors that can contribute to their success, from biotech rationale, indication selection and the clinical development setup. Not all investors are going to be able to evaluate these indicators and assess the risk/reward profile. As such, these bets are better suited to specialized life science, biotech or healthcare funds.

Approach two: Specific SaaS products

The second approach sees companies focusing on delivering SaaS products for a hyper-specific part of the drug discovery process. The challenge here is that you’re trying to target a really narrow segment of researchers. For example, out of the 2,000 researchers at AstraZeneca, only 200 focus on protein design.

While this is a good example of an interesting play, there are three ways it’s likely to pan out:

1. The startup sticks to its SaaS solution. However, in our experience, the pharma industry has a hard psychological cap on how much it’s willing to pay on tooling its researchers. This makes it a hard sell for startups trying to charge over €1m a year (the minimum amount we believe is required to build a generation-defining company) for a tool that only a handful of researchers are going to use.
2. Facing this and struggling to scale, we’re seeing many VC-backed companies (some with $50m+ revenue) opting for a success-fee model, recognising that the value is in the drug and not the software. This means revenues are no longer SaaS-driven and closer to a “classic” biotech play.
3. To build a sustainable life sciences software business, we believe startups need to address multiple scientists. To achieve this, we’re seeing some startups that set out on a specific mission eventually expand into adjacent fields connecting multiple solutions. This path tends to be very long.

What we want to see

Despite the ongoing transformation in R&D, the software just hasn’t kept up. As a result, we expect to see a new category emerge: software that, from the get go, acts as a unified AI-powered workbench for scientists, allowing them to collaborate and find insight around targets and candidates, and leverage LLM search through omics data, external papers, internal data and AI-powered simulations.

The future of R&D is networked and collaborative, and these new vendors will facilitate better and more efficient utilization of data and AI across entire organizations. This will provide super powers to everyone on top of their current streams, while allowing them to break internal knowledge silos and organizational boundaries.

We believe such platforms could help accelerate the impact of AI in understanding and modifying biology — a requirement to improve targets and increase personalisation of medicine.

If you’re building in this space, whether you’re already up and running or just thinking about getting started, please get in touch.

Think we’re wrong? Please get in touch.