AI is accelerating clinical development – however, discovery is slower to benefit owing to sparse, siloed data, meaning human expertise still matters enormously.
Dr. Henning Steinhagen, a senior executive with 25 years’ experience in the pharma, biotech, CRO/CDMO industries and VC, spoke at Inflexion's most recent annual Pharma Services dinner. The dinner brought together leaders from across the sector to discuss the emerging trends shaping the industry.
Following the event, Henning expanded on the thinking he had shared regarding AI’s potential in the sector.
What are the most material ways AI is changing pharma services today?
There are basically three phases to consider: target identification (finding biological pathways worth interrogating), drug discovery (choosing a modality – small molecule or biologic – and designing candidates), and drug development (clinical trials through to approval). AI is now supporting in all three, with the biggest past and near-term gains in development.
Upstream, structure determination is a good example of AI boosting experimental science. Techniques such as X-ray crystallography and cryo-EM have transformed how we get 3D structures of complex proteins today; AI tools like AlphaFold3 can now augment those outputs by refining models and speeding interpretation. It’s all about combination: experimental data plus computational prediction to guide medicinal chemistry faster. CROs that specialise in structural biology are driving this forward by showing how a drug binds – then using those insights to iterate for better candidates.
Downstream, AI is already embedded in clinical development, and it improves each year in areas such as trial design and simulation, patient stratification and site selection, biomarker discovery, and safety monitoring. Real-world examples include platforms that pre-screen electronic records to identify eligible patients and trial designs that use fewer, but more relevant, participants – improving timelines and cost.
Several emerging technology-driven companies have shown how curated, analysable data improve both speed and quality. That momentum compounds as more outcomes are published and reused to train better models.
Has AI helped drug discovery and development?
AI is omnipresent, but far from being omnipotent. There are areas where impact is clear, and others where promise runs ahead of proof. Drug R&D remains to be a decade-plus journey with high attrition and total costs of over $2bn per approved drug. Along that path, AI already adds value – but where success depends on access of high-quality data (especially in discovery), impact is still limited.
And there are limitations. A lack of relevant and available discovery data means that AI can’t build and train relevant models in order to make meaningful impact on drug discovery yet. Especially in the preclinical stage key data and knowledge (i.e. on SAR, ADME, PK) is mainly sitting in individual companies’ notebooks, spreadsheets and scientists’ heads – so for now, even the best models are nothing more than algorithms as they can’t access large parts of this valuable data pool. That’s why in contrast, models like AlphaFold3 are so impressive and impactful: they solved a well-framed problem with abundant training material. But medicinal chemistry, which is messier as it is complex and multi-objective, remains a very hand-crafted art. There are already some useful programs emerging but true impact of AI on small molecule drug discovery and design will take many years to come. Therefore, I don’t expect a wholesale replacement of medicinal chemists any time soon.
In the long run, future generations of scientists and chemists need to remain critical but also embrace and use new AI-based technologies in order to become more efficient and successful.
Where are we seeing real productivity gains and what is hype?
Discovery is where the reality lags the rhetoric. There are headline claims of “AI-discovered” drugs, but many are re-purposings or heavily human-guided programmes. We’re not yet able to ask a machine to design a drug from scratch and it will be years to come.
In contrast, development has the data to train robust models using government listings for clinical trials, published outcomes, real-world evidence, and so forth. That’s why tools for trial design, patient identification and stratification have scaled already with success.
How should CROs structure their use of AI to maximise impact?
Apart from experimental techniques, they first and foremost need access to relevant data to develop prediction powers that can be used to turn data into information. CROs also need to better understand the ‘customer journey’ of drug discovery and development in order to become real partners and more than just capacity providers of a specific tool or technique.
For example, structure-first CROs should pair their experimental engines (crystallography, cryo-EM) with model-driven guidance that answers practical questions: Which modifications are most promising? What’s the fastest route to a better candidate? In other words, giving scientists not just the data but a push on how they can better interpret it. That requires building (or partnering for) computational capability and integrating it into workflows.
Europe has a strong and growing AI talent base, but the market is fragmented; critical mass often requires alliances – CROs teaming up with AI specialists and pharma professionals – rather than trying to build everything organically and in house. Differentiation will ultimately come from combining wet-lab excellence, computational power and real success stories.
What realistic progress should we expect in discovery?
I hope that one day AI can truly help us to design and select drugs candidates with overall superior properties, but this is a complex multi-dimensional game and we’re not there yet. We can reasonably expect useful tools: structure prediction will keep helping, generative models will propose more relevant chemistries, scoring functions will get sharper, and self-learning algorithms will help to get better compounds faster.
Ideally, we will eventually see a system that proposes only a handful of predicted preferred structures with quantified trade-offs so a chemist can make a better-informed judgement and significantly shorten the discovery timelines.
How is private equity helping pharma-services businesses embrace or accelerate AI?
It is important not only to act as money providers, but also to add value by bringing deep industry knowledge. PE can act as a very astute sounding board as they ask management to define its AI strategy, helping to refine recruitment and then execute to ensure it’s the right people in the right place for the journey ahead. They can also help with M&A to accelerate the journey by combining a computational shop with a wet-lab platform (or vice-versa) to create the dual capability the market now rewards. Buy and build is a big area of expertise with PE and they bring deep understanding and ideally a broad adviser network.
Bottom line is that by investing in the right businesses PE can play a key role in accelerating AI-based technologies to become more impactful for the benefit of patients in dire need of new therapies.
Ambitious businesses can grow faster with the right capital and expertise, with Inflexion’s flexible funding offering minority or majority capital and access to a sizable team to accelerate growth. The Inflexion team has significant experience in supporting the growth of a number of healthcare companies of different sizes, with recent successes such as ELCG and Medivet, and a growing portfolio which includes CNX Therapeutics, Ensera, Proteros, Rosemont Pharmaceuticals, Tierarzt, Upperton Pharma Solutions and Village Vets. All have access to Inflexion’s value acceleration strategies of M&A, digital enhancement, international expansion, commercial effectiveness, sustainability and talent management.
