Harnessing AI's potential: Empowering researchers to accelerate drug discovery

Large language models (LLMs) are becoming essential to how life sciences organizations discover and develop new therapies. In a recent webinar, Chris McSpiritt, VP of Life Sciences Strategy at Domino, shared how LLMs are reshaping drug discovery.

The evolution of LLMs

While LLMs have been in use for decades, the past few years have seen a significant increase in capabilities and adoption, driven by advances from open source projects like Meta’s LLaMA and commercial models like GPT-4 and Claude. Today, LLMs generate text, images, and even working code to support complex scientific workflows.

LLMs across the drug development lifecycle

But LLMs aren’t replacing scientists and programmers — they’re amplifying their capabilities. Here are a few examples of how life sciences companies use LLMs throughout R&D.

Bioinformatics and drug discovery

LLMs can help researchers synthesize findings from thousands of publications, speed literature reviews, predict genomic and protein sequences, generate novel CRISPR edits, and design new molecules. These capabilities free up time for hypothesis testing and experimentation.

Preclinical research

Scientists can use LLMs to predict how new compounds behave, modeling toxicity, absorption, and other pharmacokinetic properties. This means less iterations in labs and faster progress toward clinical trials. McKinsey estimates AI could reduce this stage from months to weeks.

Clinical trials

Clinical development is an area where LLMs can have a big impact. A few examples are:

• Drafting research plans and documentation based on high-level scientific goals
• Analyzing prior trial designs and amendments to optimize trial design
• Suggesting inclusion/exclusion criteria to improve enrollment
• Mining electronic health records (EHR) to identify eligible patients and identify early signs of dropout or adverse events

When every day can cost $500,000 in unrealized or lost prescription drug sales and $40,000 in direct daily clinical trial costs, these efficiencies can drive significant savings.

Statistical programming and submissions

LLMs can accelerate timelines by generating statistical code, validating outputs, and authoring clinical study reports (CSRs). LLMs pull data from protocols and statistical analysis plans to streamline the CSR drafting process and reduce submission bottlenecks.

Corporate operations

While many start with individual adoption, such as scientists using ChatGPT to write summaries and emails, forward-thinking companies like Moderna are going all-in, licensing tools like ChatGPT for use enterprise-wide. LLMs can help automate workstreams in HR, finance, and IT, freeing up time for more high-value tasks.

Keys to success

Think of LLMs like exceptionally bright and energetic interns — they’re fast, eager, and completely capable, but they still need guidance, context, and a manager to keep them on track, so a human in the loop remains critical. Companies also need to align processes and technology to realize the full value of LLMs by:

• Fostering a culture of collaboration by eliminating silos between data scientists and business teams
• Providing clear governance by defining how and where LLMs should be used
• Leveraging platforms like Domino to fine-tune models and control access

Agentic AI and unified data

The future lies in agentic AI: LLMs capable of independently planning and executing tasks rather than simply responding to prompts. Having a virtual employee draft a protocol and refine it, versus writing it from scratch, is not far off. But this shift requires life sciences organizations to unify data across functions, so AI can derive insights from bioinformatics, clinical operations, and safety. Retrieval-augmented generation (RAG) will be integral to query systems and provide context to outputs.

Looking ahead

LLMs are no longer emerging — they’re here to stay, and they’re transforming how companies develop drugs and run clinical trials. The opportunity lies in the ability to build the right culture, controls, and infrastructure to scale safely and effectively. As McSpiritt said, “LLMs aren’t here to replace us. They’re here to assist us and give us a head start.”

Join 150+ data science and IT leaders at RevX in Philadelphia on May 20 to dive deeper into practical uses for LLMs and hear best practices from companies like AstraZeneca.