Capgemini Q&A: AI's Role in the Future of Drug Discovery

What parts of drug discovery do you think AI will transform most significantly, and is the bigger opportunity speed or improving the quality of drug candidates?

For me, the biggest opportunities are in target discovery, hit and lead prioritisation, and in silico experimentation that can reduce reliance on some forms of laboratory and animal testing. And this is supported by what we see in our recent Capgemini Research Institute report where 63% of biopharma executives believe the majority of new molecular entities (NMEs) will originate from AI-driven platforms within the next decade.

AI can certainly accelerate parts of the discovery process, but I think the more significant impact is on quality. Historically, drug discovery has involved screening enormous numbers of potential drug candidate molecules and accepting a high degree of failure. AI gives us the ability to identify more promising targets faster and design candidate molecules far more efficiently.

For example, Insilico Medicine’s average timeline to nominate a preclinical candidate is between nine and 12 months vs the industry average of four to six years. Increasingly, the biopharma industry is seeing improved probability of success in phase one trials indicating higher quality candidates are coming through from AI. 

What's particularly exciting is that AI may also help us tackle diseases that have traditionally been considered intractable, whether through advanced simulation techniques or by identifying smaller patient populations where therapies that previously failed could actually prove effective

A good example is advancement in idiopathic pulmonary fibrosis (IPF), a severe and progressive lung disease with limited treatment options. Insilico Medicine demonstrated how AI can transform drug discovery by identifying TRAF2- and NCK-interacting kinase (TNIK) as a novel target for IPF. Insilico Medicine analysed large-scale biological data to uncover TNIK as a previously overlooked but highly relevant driver of fibrosis.

This AI-driven insight enabled rapid development of a first-in-class therapy, reducing the timeline from target identification to clinical candidate from five to seven years to under three years, Early clinical data has shown encouraging safety and efficacy signals, demonstrating how AI can not only accelerate discovery but also improve the quality of the targets and candidates entering development.

Why is it important for AI to be integrated directly into automated physical laboratories? What does the convergence between wet labs, in silico labs, and AI look like in practice?

Most organisations still think about AI as something separate from the laboratory. In reality, we see best results when AI is tightly integrated into physical experimentation. For example, Capgemini’s deep tech powerhouse Cambridge Consultants combines wet lab capabilities within in silico experimentation environments that allow organisations to define, run and refine experiments much faster and with fewer required data points. Bringing together physical experimentation, data structures and algorithms in this way creates a self-learning system that can accelerate innovation.

This becomes particularly powerful when biopharma organisations create closed-loop environments where physical laboratories, computational models and AI systems continuously learn from one another. Experimental results generated in a wet lab feed directly into AI models, which refine hypotheses and recommend the next set of experiments. AI has proven to excel in protein design where the model itself is making decisions on the unique string of amino acids for the optimal 3D structure of a specific protein.

For example, in our Cambridge Consultants labs, we were able to achieve the brightest Green Fluorescent Protein (GFP) expression variant with only 46 data points compared to an estimated 80,000 required through directed evolution. Those experiments then generate new data that further improves the models. This is particularly important because scientific research requires highly specialised models trained on domain-specific data. 

Where do you see the biggest opportunities for AI to improve clinical trial efficiency?

Our research finds that 60% of R&D leaders believe AI will substantially improve clinical trial efficiency, with opportunities across the entire clinical development process.

Currently, we see some of the most immediate benefits in patient recruitment, site selection, operational management and trial design. AI can help identify eligible participants faster, improve forecasting around recruitment for clinical trials, and optimise trial protocols for both efficacy and safety. In addition, there is significant potential in clinical statistical analysis where time can be compressed from weeks to days. 

Looking further ahead, synthetic and simulated data could even change how clinical evidence is generated. One particularly promising area is the use of synthetic control arms for severe or rare diseases, where reducing reliance on placebo groups is both scientifically and ethically compelling.

We are already seeing encouraging examples emerge. In one study, researchers successfully recreated the control arm of a randomised clinical trial using data from previous studies and real-world patient registries, achieving comparable survival outcomes to the original trial. Approaches like this could be especially valuable for patient populations that are traditionally difficult to recruit into clinical studies, including elderly patients and those with rare diseases. 

With AI, researchers could model long-term patient outcomes more effectively, reducing the need for some patients to remain on treatments for extended observational periods.

Together, these developments could reduce trial sizes, shorten timelines and accelerate access to new therapies. Importantly, before this happens researchers will need to establish sufficient regulatory confidence in the models and evidence that underpin these approaches.

What role could AI agents play in scientific research, and what are the biggest barriers to adoption in biopharma R&D?

I see AI agents as companions for scientists. Their value lies in helping researchers challenge assumptions, test hypotheses more rigorously and eliminate weaker ideas earlier in the process. According to our research, 38% of organisations are already piloting AI agents in R&D. 

Scientific research requires very specialised AI models that can explain their reasoning, justify conclusions and operate within established scientific principles. 

To unlock the full potential of AI agents, specialised models must be trained on scientific data, supported by synthetic datasets where appropriate, and grounded in biological and molecular understanding. We also need stronger data foundations, governance and trust in AI outputs. For example, an AI model may identify a promising drug target, but scientists need confidence that the recommendation is based on complete, high-quality data and can be explained and validated. Those factors are just as important as the technology itself. 

Looking ahead five to ten years, how fundamentally could AI reshape pharmaceutical R&D and the kinds of diseases the industry is able to target?

I believe AI has the potential to transform pharmaceutical R&D more than any technology we have seen in decades.

In the near term, we will see faster and more efficient research processes. But the longer-term impact is truly exciting – pursuing biological targets that were previously considered too complex, identifying new patient populations for existing therapies, and dramatically improving our understanding of disease mechanisms. 

As those capabilities mature, AI will move from being a tool that supports R&D to becoming a core component of how new medicines are discovered, developed and validated. Ultimately, that could enable the industry to make progress fighting diseases that have historically proved difficult to address, including complex neurological disorders, rare diseases and highly heterogeneous cancers, improving outcomes for patients around the world.

Fundamentally, AI is rewriting how we approach the assessment of targets and what is technically feasible in terms of what diseases we can treat. There are some highly ambitious targets here such as claims to target solving all disease. That remains to be seen, but the move toward disease and modality agnostic AI platforms creates an opportunity not previously seen in life sciences.