AI Improves R&D and Reduces the Risk of Failure
Identifying new molecules, anticipating their toxicity, predicting the outcomes of clinical trials… Artificial intelligence is already transforming pharmaceutical research. By 2035, it could accelerate the development of more effective and personalized treatments, while reducing the risk of failure.
Developing a new drug is a long, costly, and uncertain process. To achieve a conclusive result, it takes between 15 and 20 years of research and nearly 2 billion euros in investment, all while dealing with the risk of failure at every stage. AI could be a game-changer by enabling researchers to more quickly select promising molecules, anticipate their adverse effects, and better predict their clinical efficacy.
Algorithms are already analyzing vast amounts of biological, chemical, and medical data to identify therapeutic targets and model interactions between molecules and proteins. Generative AI also makes it possible to design new molecular structures. Rentosertib, a drug currently in development for idiopathic pulmonary fibrosis, illustrates this trend. Developed with the help of AI, it completed Phase III clinical trials in 2026—a first. It is scheduled to be launched in 2030. We discuss this with Frédéric Girard, president of France Biotech, an association that brings together innovative HealthTech companies.
The regulatory framework must evolve without compromising patient safety.
Frédéric Girard
president of France Biotech
Four Scenarios for Pharmaceutical Research in 2035
In the future, following the example of rentosertib, models could better anticipate a formulation’s toxicity and eliminate the least promising candidates early on in the preclinical phase. Further down the pipeline, they could help optimize clinical trials and identify the populations most likely to respond to a treatment.
For Frédéric Girard, the economic benefit lies primarily in risk reduction: “If we can identify very early on the molecules with the greatest potential for efficacy and the highest likelihood of safety, this can significantly reduce the risks associated with development and reassure investors.”
This trend will not necessarily result in cheaper drugs, particularly because regulations are becoming more stringent and licensing agreements will continue to weigh on their value. However, it could improve the allocation of investments and open up therapeutic areas that are currently neglected due to the uncertainty surrounding them.
Scenario 1: AI Becomes an Indispensable Co-Pilot for Researchers
The first scenario, which most closely aligns with current trends: AI is gradually integrated into the entire research and development chain without disrupting its organization. Algorithms become decision-support tools for identifying biological targets, generating molecules, predicting their toxicity, selecting experiments, and preparing clinical trials.
Researchers retain a central role in formulating hypotheses, interpreting results, and validating them experimentally. AI does not replace biological research, but it allows resources to be focused on the most relevant avenues. This trajectory, however, requires continuous improvement in the reliability of models and widespread adoption of digital tools.
Scenario 2: Drugs Tailored to Specific Patient Profiles
Second pathway: The convergence of AI, molecular biology, and health data accelerates the rise of precision medicine.
By simultaneously analyzing thousands of genetic, biological, and clinical parameters, algorithms could identify subpopulations of patients with distinct pathological mechanisms. Certain diseases currently considered single entities could thus be subdivided into several forms requiring different treatments.
As Frédéric Girard explains, depression illustrates this perspective because a single diagnosis can correspond to twenty or thirty different clinical profiles, each associated with varying responses to antidepressants. AI would help better identify patients likely to benefit from a given drug, thereby avoiding the indiscriminate testing of the same treatment on heterogeneous populations.
The logic of pharmaceutical development would then change. It would no longer be merely a matter of finding a drug for a disease, but of determining which treatment is best suited to which group of patients.
Scenario 3: Research processes largely driven by predictive models
A third, more transformative possibility involves multimodal AI systems that combine molecular, genomic, clinical, and experimental data to build increasingly precise representations of biological mechanisms.
When paired with automated laboratories, these models could fuel continuous cycles of design, experimentation, and learning. Each result would enrich the predictions and guide subsequent experiments. Research would thus evolve into a semi-automated, iterative process capable of rapidly testing numerous hypotheses.
This approach could accelerate the discovery of molecules, better predict their pharmacological activity, and limit experimental dead ends. However, it would remain dependent on the quality of the data used, the models’ ability to represent biological complexity, and the scientific validation of the results.
Scenario 4: Radically Reconfigured Clinical Trials
Fourth scenario: AI is also transforming the way drugs are tested in humans. Simulations and “virtual patients” could help compare different protocols before they are launched. For certain rare diseases, AI could also help construct synthetic control groups based on data describing the natural progression of the disease.
This issue is particularly significant in the case of serious and irreversible pediatric conditions, where administering a placebo can pose major ethical challenges. Patients receiving the experimental treatment could, in certain situations, be compared to a modeled progression of the disease rather than to a traditional placebo group. However, this approach will depend on acceptance by health authorities.
As Frédéric Girard points out: “The regulatory framework must evolve without compromising patient safety.”
These approaches will need to demonstrate their scientific robustness, particularly to avoid biases linked to historical data and to ensure a sufficient level of evidence. The transformation of clinical trials will therefore depend as much on regulatory advances as on technological progress.
Access to Data: A Decisive Factor in the Shift Toward AI
Regardless of the scenario, data availability is a determining factor. To design and train their models, companies will need information from molecular chemistry, biological research, clinical trials, and routine care.
However, in France and Europe, this data remains scattered across heterogeneous systems that are often not very interoperable. Access to it remains complex, even when it could be used for research in compliance with personal data protection rules. Frédéric Girard highlights this challenge: “Major French biotech companies are sourcing their data from the United States. As a result, we are missing out on a major economic opportunity. ”
This dependence also raises questions of sovereignty and representativeness, as U.S. data does not necessarily reflect all the characteristics of European populations. By 2035, the creation of secure and interoperable health data repositories could therefore become a major driver of competitiveness for European pharmaceutical research. Without easy access to reliable and representative data, even the most effective models risk being limited in their applications.
From “Digital Pharma” to “Digital Care”
Another consequence of this acceleration could involve the financing of innovations. While AI increases the number of promising drugs, healthcare systems will need to find ways to make them accessible to patients.
Frédéric Girard identifies two complementary trends: “digital pharma,” which accelerates and ensures the safety of treatment development, and “digital care,” which improves the efficiency of care pathways. In practical terms, better medical coordination, shorter hospital stays, enhanced home care, and a reduction in avoidable medical interventions could generate savings while improving the patient experience. These gains would thus help finance the introduction of new therapies.
By 2035, AI could thus become a fundamental component of pharmaceutical research, from drug discovery through clinical trials to personalized treatments. The pace of this transformation will, however, depend on the reliability of models, access to data, regulatory developments, and funding for innovation.
The major breakthrough could come from the ability to integrate research, data, and healthcare delivery. AI will not eliminate biomedical uncertainty, but it could significantly reduce it, to the benefit of patients.
This text has been translated by an artificial intelligence.
Glossary
Generative AI
Artificial intelligence technology capable of creating new content based on training data. In pharmaceutical research, it can, in particular, propose novel molecular structures with therapeutic potential.
Drug candidate
A substance identified as having therapeutic potential and undergoing evaluation prior to potential marketing authorization.
Preclinical phase
A stage of research preceding human trials, intended in particular to study a molecule’s biological activity, toxicity, and behavior.
Clinical trial
A study conducted in humans to evaluate the safety, efficacy, and adverse effects of an experimental treatment.
Predictive model
A statistical or algorithmic system that uses data to estimate a future outcome, such as a molecule’s toxicity or a response to treatment.
Precision medicine
A medical approach aimed at tailoring treatments to the biological, genetic, and clinical characteristics of patients or groups of patients.
Virtual patient
A computer model of a patient or patient profile used to simulate the progression of a disease or the possible effects of a treatment.
Synthetic control arm
A comparison group constructed from existing data, particularly historical data, to estimate what might occur in the absence of the experimental treatment.
Data interoperability
The ability of different computer systems to exchange data and use it consistently, particularly in the healthcare sector.
We quote them
Frédéric Girard
Frédéric Girard has served as president of France Biotech since 2024, the association that brings together French entrepreneurs in healthcare innovation. A doctor of medicine with a degree in business administration, he currently serves as Senior Vice President of European Development Operations at DBV Technologies.
Throughout his career spanning more than 30 years in the healthcare industry, he has held executive positions at pharmaceutical companies such as Novartis, AstraZeneca, and Servier. Committed to promoting French HealthTech, he works to advance innovation funding, international development, and health sovereignty.
FAQ
AI rapidly analyzes vast sets of biological and chemical data to identify therapeutic targets, select promising molecules, and predict some of their properties. It can thus reduce the number of experiments required and help researchers focus their efforts on the most relevant leads.
No. While generative AI is capable of designing new molecular structures, these must be synthesized, tested, and validated experimentally. Researchers remain essential for interpreting results, assessing risks, and conducting clinical trials. A drug designed with the help of AI must meet the same regulatory requirements as any other treatment.
Not necessarily. By reducing certain research costs and the risk of failure, AI could improve the efficiency of pharmaceutical investments. However, the final price of a drug also depends on manufacturing costs, commercial agreements, intellectual property, and pricing and reimbursement mechanisms.
AI could facilitate participant recruitment, identify patient profiles likely to respond to a treatment, and optimize experimental protocols. Simulations and synthetic control groups could also supplement certain studies, particularly for rare diseases, subject to rigorous scientific and regulatory validation.
Artificial intelligence could accelerate precision medicine by identifying subgroups of patients who share certain genetic, biological, or clinical characteristics. The goal would be to better tailor treatments to patient profiles, though this does not guarantee the creation of a drug specific to each individual.
