AI-driven drug discovery is not just accelerating pharmaceutical pipelines but redefining how we understand, simulate, and target disease. As large language models and generative design tools mature, drug development is entering an era of computationally led precision.
The pharmaceutical sector has always been defined by its complexity. Yet today’s complexity is reaching a scale and scope where human expertise alone can no longer fully comprehend or navigate it. The convergence of biomedical research and AI is not merely automating processes but actively transforming the cognitive infrastructure of drug discovery.
Nicola Richmond, Chief Scientist AI at Recursion, believes the traditional approach is reaching its limits. “Year on year for many years now, the conventional approach to drug discovery has become increasingly costly and unproductive,” she says. “Harnessing AI offers the possibility of revolutionising the status quo, reducing the time and cost of drug discovery, but also enabling us to find better-targeted medicines for specific patient types, where efficacy and tolerability are optimised.”
Lorena Corfas, Global Media Spokesperson at Roche, sees the same shift. “AI is revolutionising healthcare,” she says. “The breakthrough of generative AI could represent another pivotal moment for the industry, comparable to paradigm shifts such as bringing molecular biology to drug discovery in the 1970s.”
While high-throughput screening and algorithmic modelling have existed for decades, the capability to generate, evaluate, and iterate on chemical entities in silico at scale is something entirely new. What distinguishes this generation of tools is not speed alone but adaptive intelligence, systems that learn from multimodal data, simulate synthesis pathways and optimise for biological relevance in real time.
Reprogramming the drug design cycle
At the core of this transition lies the growing maturity of generative AI and large language models. Richmond explains that Recursion’s approach is grounded in a full-stack pipeline that connects generative design with real-world synthesis and evaluation. “We are developing and combining AI methods to assess synthetic accessibility of the designs,” she says. “This end-to-end design-score-synthesise pipeline is a compelling approach to make drug discovery more efficient and less costly.”
This integration is not confined to molecular generation. Roche has developed what it describes as a ‘lab-in-a-loop’ system, where experimental data feeds into computational models to generate new, experimentally testable predictions. The result is an iterative, hybrid loop of experimentation and simulation that moves the drug development process from linear to cyclical.
“The potential of AI when it comes to fuelling and speeding up innovation across Pharma and Diagnostics is tremendous,” Corfas says. “Thanks to the convergence of science and technology and their transformative advances, we now have an opportunity to bring multiplicative, rather than incremental, benefits to patients.”
The scale of that impact is amplified by the nature of AI’s data appetite. AI thrives not on abstraction but on detail, on the messy, heterogeneous, high-dimensional data generated at every stage of the development cycle. In that context, biology is not an obstacle but a dataset.
Learning from the edge of knowledge
One of the most profound shifts brought about by AI in drug discovery is the ability to make informed bets in uncharted biological terrain. Richmond notes that AI allows teams to explore chemical space even when historical knowledge is limited. “Chemical space is huge, more than the number of atoms in the known universe,” she says. “Our methods allow us to generate reliable property predictions, even when we know little about the chemical space related to the disease biology we are targeting.”
This capability is significant for complex diseases with multifactorial aetiologies, such as cancers or neurodegenerative disorders, where conventional reductionist approaches fall short. It also unlocks new potential for patient stratification and personalised treatment.
Corfas points to Roche’s use of AI in real-world applications, such as its Navify Algorithm Suite, which includes AI-powered clinical algorithms to aid early disease detection and treatment planning. “Our AI efforts address opportunities at every stage of a patient journey,” she says. “We are commercialising and piloting the use of evidence-backed algorithms to assist clinical decision-making by physicians to identify, detect and manage the disease.”
Beyond diagnostics, machine learning is also being used to support patient cohort selection for clinical trials, reduce trial attrition, and simulate molecular dynamics in silico. Each advance not only reduces cost but improves the fidelity of the therapeutic match between molecule and patient.
Challenges and responsibility in the loop
For all its promise, AI in drug development is also bound by the same regulatory and ethical frameworks that govern the broader healthcare industry. This includes issues of bias, data privacy, explainability, and the need for human accountability.
Corfas stresses that Roche is deeply engaged with regulatory bodies and maintains strong ethical guardrails in its deployment. “AI is and will be a highly regulated tool in our toolbox,” she says. “We assess the potential risks that may result from the use of AI systems and take measures to mitigate these risks during both the development and deployment of AI solutions.”
These safeguards include human-in-the-loop systems, bias auditing, and anonymisation of training data to prevent the reidentification of individual patients. Similar principles apply at Recursion. “Whether or not a company uses AI to identify a drug candidate, we are still bound by the same regulatory framework for data privacy,” Richmond says. “The industry, including Recursion, goes to great efforts to ensure personal data privacy is protected.”
Ethics in AI is not limited to patient protection; it also covers inclusivity in training data, equity in access, and transparency in algorithmic decision-making. There is growing awareness across the sector that trust will determine the future scalability of AI-driven drug development, not just technical capability.
Collaboration as infrastructure
If AI represents a shift in how drugs are discovered, it also demands a new operational model for how that discovery takes place. Neither Recursion nor Roche position themselves as isolated innovators. The success of AI in life sciences increasingly depends on a mesh of partnerships across technology, academia, healthcare, and pharma.
“Collaboration with other players across industry and academia is a critical factor in the success of the life sciences ecosystem,” Richmond explains. “Pharmaceutical companies provide much-needed investment and capabilities into researching new treatments in areas of high unmet need. Through their investment, pharma companies also help shoulder some of the risk involved in early drug discovery.”
Corfas reinforces that perspective. “We harness external innovation to complement our in-house expertise,” she says. “By following the science, utilising cutting-edge technologies, and strategic partnerships, we aim to shape the future of AI in drug development and engineer the foundations of tomorrow’s discoveries.”
There is now an emerging consensus that AI is not a tool to be bolted onto old ways of working. It is a paradigm shift that requires rethinking how institutions share data, validate findings, and iterate toward clinical success.
Tomorrow’s therapeutic landscape
The long-term impact of AI on drug discovery will not be defined by one platform, model, or company but by a structural shift in how biological understanding is converted into medical interventions. Richmond sees particular promise in LLMs as agentic systems, able to synthesise knowledge and imagery to support decision-making across multimodal data. “Much of the data we generate is multimodal,” she says. “With an ever-increasing data landscape, we will rely on powerful models to help us understand this data and decide on what is the best path forward.”
Corfas is equally clear-eyed about the opportunity. “We are at an inflexion point in the history of healthcare,” she says. “We are working every day at the convergence of technologies, platforms, biological understanding and data science, all in the service of patients.”
Whether that future involves digital twins of patient cohorts, AI-driven adaptive trials, or fully automated pipelines from in silico design to human testing, one thing is clear: medicine is becoming computational. And those who learn to think in code and molecule alike will shape the therapies of the next century.
