TLDR: A study by Weave Platform and Takeda Pharmaceuticals found that an AI-powered platform, AutoIND, can reduce the time to draft initial Investigational New Drug (IND) applications by approximately 97% compared to manual methods. While the AI-generated drafts achieved acceptable quality with no critical errors, human experts are still required to refine the content for submission-readiness, particularly in areas like conciseness, emphasis, and completeness, due to systematic AI deficiencies. The research advocates for a human-AI collaborative model to enhance efficiency and consistency in regulatory writing.
The pharmaceutical industry is constantly seeking ways to accelerate drug development, and a recent study highlights how artificial intelligence (AI) can significantly streamline one of its most time-consuming processes: preparing Investigational New Drug (IND) applications. A collaborative research effort between Weave Platform and Takeda Pharmaceuticals has demonstrated that human-AI teamwork can dramatically boost efficiency in regulatory writing, specifically for the initial drafts of IND submissions.
The Challenge of IND Applications
Bringing a new drug from preclinical research to human clinical trials is a complex and lengthy journey. A critical step in this process is the submission of an Investigational New Drug (IND) application to regulatory bodies like the U.S. Food and Drug Administration (FDA). These applications are massive, often thousands of pages long, compiling extensive preclinical study reports, manufacturing details, and clinical protocols. Traditionally, compiling these nonclinical written summaries (eCTD modules 2.6.2, 2.6.4, 2.6.6) is a heavily manual, expertise-dependent, and time-intensive task, frequently becoming a bottleneck that delays the start of clinical trials.
AI Steps In: The AutoIND Platform
Recognizing this challenge, the researchers aimed to evaluate whether a large language model (LLM)-based platform, named AutoIND, could reduce the time required for first-draft composition of IND applications while maintaining document quality. AutoIND, integrated into The Weave Platform, leverages advanced AI to generate draft IND sections from various source documents.
Remarkable Time Savings
The study’s results on time efficiency were striking. AutoIND reduced the initial drafting time for nonclinical written summaries by approximately 97% compared to traditional manual methods. For instance, for one IND application (IND-1) involving 61 source documents and 18,870 pages, AutoIND generated a complete first draft in just 3.7 hours, a task that typically takes an estimated 100 hours manually. Similarly, for another IND (IND-2) with 58 documents and 11,425 pages, the AI completed the draft in 2.6 hours, again against an estimated 100 hours for manual preparation. This translates to a significant increase in drafting speed, from an average of 0.2 pages per hour manually to 12.1 pages per hour with AI assistance.
Quality and the Need for Human Expertise
While the speed gains were impressive, the study also provided crucial insights into the quality of AI-generated content. An experienced regulatory writing assessor evaluated the drafts across seven quality categories: correctness, completeness, conciseness, consistency, clarity, redundancy, and prominence/emphasis. The overall quality scores for the AI-generated drafts were 69.58% for IND-1 and 77.85% for IND-2, indicating that about 10-25% of the content would require revision or refinement.
Crucially, no critical regulatory errors—such as misrepresentations or omissions that could alter regulatory interpretation of safety or efficacy—were detected. Correctness was consistently high, exceeding 89% across all modules. However, the AI exhibited systematic deficiencies, particularly in “emphasis,” often over-describing methods and under-weighting critical findings. Other areas needing refinement included conciseness (AI drafts were 3-5 times longer than human-written documents), completeness (missing essential study design elements or critical data), and clarity (structural issues and repetitive, AI-characteristic language).
These findings underscore a key takeaway: while AI can handle the systematic compilation and organization of technical data, human experts remain essential for refining the output to submission-ready quality, focusing on strategic interpretation, ensuring completeness, and correcting nuanced terminology. The systematic nature of these AI limitations, however, provides a clear roadmap for future improvements through specialized fine-tuning and structured generation templates.
Also Read:
- AI Drug Discovery’s Slow Pace: Biological Complexity and Data Challenges Beyond Technological Hurdles
- Enhancing Clinical Reasoning in AI: A Compartmentalized Approach for Language Models
A Collaborative Future for Regulatory Writing
The study champions a human-AI collaboration model, where AI accelerates the initial drafting, freeing human experts to focus on higher-value tasks like ensuring emphasis, completeness, and overall narrative consistency. This approach not only offers unparalleled speed but also helps maintain a more uniform voice and style across large documents, which can be challenging when multiple human writers are involved.
Furthermore, the research suggests that AI tools like AutoIND could serve as valuable training aids for junior medical writers and help established companies maintain consistent quality across teams. As regulatory bodies, including the FDA, increasingly move towards AI-enabled review processes and data-centric submissions, AI-driven authoring tools are well-positioned to support this evolution, provided their current challenges are addressed.
This research, detailed in the paper Human-AI Collaboration Increases Efficiency in Regulatory Writing, provides compelling evidence for the responsible adoption of AI in highly regulated industries. It highlights that while current AI systems excel at generating surface-level drafts with impressive efficiency, achieving truly submission-ready regulatory documents will require continued investment in technical fixes, structural improvements, enhanced training, and robust quality control systems.
