Accelerating R&D Pipeline With LLMS And Generative AI In Pharma: Opportunities & Common Pitfalls

5 Opportunities for LLMS and Generative AI in Pharma R&D

Opportunity n°1: Increasing Efficiency in Drug Discovery

As recent partnerships have shown (e.g., Isomorphic Labs with Novartis and Lily¹Isomorphic labs. (s. d.). https://www.isomorphiclabs.com/articles/isomorphic-labs-kicks-off-2024-with-two-pharmaceutical-collaborations), Protein Language Models (PLMs) and GenAI have the potential to transform the drug discovery process. Traditional methods require vast amounts of time and resources, but with AI-powered models, it is possible to sift through millions of chemical compounds, predict drug interactions, and even propose new molecular structures. The efficiency gained through these AI-driven models can shave years off the development process.

Opportunity n°2: Finding Opportunities in Your LIMS

Laboratory Information Management Systems (LIMS) have long been the backbone of data management in research. Integrating next generation AI models, like LLMs and VLMs, can unlock insights buried in your LIMS, helping teams identify valuable research assets more efficiently. This integration also allows to streamline the decision-making process for asset selection²Leveraging Generative AI in Laboratory Businesses : Integration with LIMS and Customer Portals | Sednor. (s. d.). https://www.sednor.io/articles/leveraging-generative-ai-in-laboratory-businesses-integration-with-lims-and-customer-portals.

Opportunity n°3: Accelerating Patient and Center Recruitment for Clinical Trials

Recruiting the right patients and centers for clinical trials is one of the most challenging aspects of clinical research. LLMs can scan through patient records, identify eligible candidates, and recommend trial sites based on historical performance and patient demographics. By accelerating recruitment, AI can reduce delays in clinical trials, bringing therapies to market faster.

Find out how our team can support you in digitizing your clinical trials >

Opportunity n°4: Integrating In Silico Medicine and Digital Twins

The advent of in silico medicine and the use of digital twins—virtual replicas of patients—present another exciting opportunity. Digital twins allow for the simulation of treatments and interventions, enabling researchers to test drug efficacy in a controlled, virtual environment.

Opportunity n°5: Easing Post-Marketing Management and Long-Term Effects

Post-marketing surveillance is critical for understanding long-term drug effects. LLMs and multimodal models can assist by continuously analyzing patient data, identifying adverse events, and predicting long-term outcomes. This real-time analysis could help companies more quickly address any emerging issues and, if needed, proactively collaborate with the medical community to make necessary changes to treatment protocols.

What are the challenges related to the use of LLMS and generative AI in pharma R&D?

Bias, Hallucinations, and Lack of Explainability

One of the primary challenges in deploying LLMs and GenAI in R&D is the risk of bias and hallucinations—where the model generates false or misleading information. Healthcare data is often incomplete or skewed, and if the AI models are not properly fine-tuned, they can perpetuate these biases. Flawed models can lead to incorrect conclusions, potentially endangering patients and creating legal liabilities for companies. AI engineers have suggested that synthetic data—artificially generated datasets that mimic real-world data— could be used as a complement to these imperfect training datasets. However, concerns were immediately raised about how using synthetic data could ultimately lead to amplified biases instead, with new models learning from data generated by originally skewed models.

Finally, the complexity and opacity of some AI systems, often referred to as “black-box” algorithms, make it challenging for clinicians to fully explain AI-driven decisions to patients.

Without clear explainability, building trust among researchers, regulators, and patients becomes a significant hurdle.

Intellectual Property and Chain of Custody

The use of GenAI introduces complexities in managing intellectual property (IP) and ensuring the proper chain of custody for data. With AI-driven insights, it is often unclear who owns the output—the organization, the AI vendor, or the developers of the AI model. Furthermore, tracing the origin and modifications of data through AI-driven processes is critical to maintaining legal and regulatory compliance. Failure to establish a clear chain of custody could lead to IP disputes and jeopardize the integrity of research findings.

Data Privacy and Security Threats

Healthcare data is one of the most sensitive and regulated types of information. The integration of GenAI into R&D workflows amplifies the risk of data breaches and unauthorized access. As AI systems process vast amounts of patient data, ensuring that this data is securely stored and transmitted is essential. Any lapses in security could result in devastating privacy violations, financial penalties, and a loss of trust among patients and stakeholders. A potential solution to this is adopting decentralized systems.

Ethical and Trust Concerns: Patient Consent

Gaining patient consent for AI-driven analyses remains a grey area. The complexity and opacity of these AI systems, with some relying on “black-box” models or third party services, complicates the informed consent process. Additionally, the sensitivity of the data utilized, such as electronic health records and genetic information, amplifies the importance of these concerns, as individuals might not fully comprehend or agree to the extent of usage and ownership of these very critical data. Addressing these issues requires developing robust consent mechanisms that ensure patients are adequately informed about how AI technologies influence their healthcare decisions, and that ensure doctors are properly trained to use these tools, address these concerns and implement procedures to guarantee data protection.

Regulatory Compliance

Healthcare regulation requires AI models in R&D to comply with strict rules on patient safety and data integrity. Companies rushing AI adoption without ensuring regulatory compliance risk delays, fines, or product withdrawals. To navigate these challenges, many firms present AI tools as “assistants” rather than medical devices, allowing for faster market entry under less stringent regulations. However, as AI plays a greater role in clinical decisions, regulators are tightening oversight to prevent unregulated tools from impacting patient care. In response, the FDA and EMA are developing new frameworks to balance innovation with safety. Initiatives like the European AI Act or FDA’s Software as a Medical Device (SaMD) framework highlight the evolving approach to AI governance in healthcare.

The rapid adoption of LLMs and GenAI in healthcare R&D presents both opportunities and risks. While the potential to streamline processes and generate new insights is exciting, companies must tread carefully to avoid the pitfalls of bias, data security breaches, and regulatory challenges. Navigating these complexities requires expertise in AI and the healthcare industry stakes, as well as a strategic approach to implementation. Nautilus.ai, Alcimed’s Data & AI team is here to assist your organization in overcoming these hurdles, ensuring that your AI-driven R&D pipeline delivers on its promise without backfiring. Should you want to discuss a project, don’t hesitate to contact our team!

About the author,

Matthieu, Manager at Nautilus.ai, Alcimed’s Data & AI team in USA.