JPM ‘25 From the Trenches: AI in Biotech Gets Tactical

Note: we know it’s been a chaotic week in research, given the federal government’s announcement of a pause in all federal grant funding (here’s coverage from C&EN). Though that memo has since been rescinded, we realize a lot of folks are intensely uncertain about what will happen.  If you use our tools and need accommodations because of the situation, please feel free to reach out to us at here.

JPM ‘25 catches us at a time of greater flux than any since COVID-19. 

The name of the game over the last week wasn’t AI that “solves biology” but AI for researchers and drug discovery leaders that enables them to focus on the most important parts of their jobs. This includes streamlining complex workflows, reducing approval times, and connecting researchers with sophisticated tools through easier interfaces. A common thread was that AI is really only as impactful as the portion of the drug discovery workforce you can use it for.

With the new U.S. administration under President Trump, we’re all are anticipating potential shifts in healthcare policies, particularly around drug pricing reforms and regulatory approvals. There will be good, bad, and ugly – all of which was on tap to discuss at JPM.

Here were our takeaways on the ground…

Russ Altman on AI and regulatory submissions

In one talk we attended at the CERSI Summit, Russ Altman - professor of bioengineering, genetics, medicine, and biomedical data science at Stanford - highlighted the development of AI assistants designed to aid the FDA in regulatory submissions, aiming to streamline complex workflows and reduce approval times. This move was reflected in many of the conversations we had.

With that said, Altman highlighted some challenges to be on the lookout for.  

• Math woes: Large language models (LLMs) often struggle with complex mathematical computations essential for drug development, such as determining maximum tolerated doses and half-lives. 
• Tabular troubles: LLMs can have trouble when working with numerous tables (such as in the case of regulatory applications). Sidenote: We think there’s a big opportunity here for harnessing non-deterministic systems in a way where they stick to context in cases like this.
• Vocab quiz: The technical vocabulary regulatory documents pose significant hurdles for interpretation by  out-of-the-box AI. Though not in a regulatory context, this is where we see interfaces like Balto coming into play, AI custom trained around specialized tooling and terminology for a “last mile delivery” to researchers.

Despite these obstacles, the industry remains optimistic, viewing AI as a tool to augment human capabilities rather than replace them. As Janet Woodcock, former FDA principal deputy commissioner, noted, AI should be seen as a substitute for human busywork, not human judgment. With this said, Altman did note he envisions a future where AI models could both write and review future applications, prompting reflection on what precisely humans’ role in the process will be.

You can watch the full CERSI Summit talks here (click on the playlist in the embedded Youtube video to find the panels on AI, debates about transparency on regulatory decisions, and more)!

OpenAI’s Rahul Arora's on benchmarking

Rahul Arora, a researcher at OpenAI specializing in AI and Bio-AI policy, emphasized the critical need for responsible AI deployment in high-stakes fields like regulatory science. He introduced several benchmarking tools designed to evaluate and enhance AI performance in specialized tasks:

• Graduate Benchmark: This tool addresses complex, expert-level queries that are not easily searchable online, providing a means to assess AI's ability to handle nuanced information.
• SWEbench: Aimed at evaluating software engineering tasks, SWEbench offers a framework to measure AI proficiency in coding and debugging, reflecting the growing integration of AI in software development.
• Elicit: An AI tool that automates various aspects of systematic reviews, Elicit enhances efficiency in data analysis by streamlining the synthesis of research findings.

Arora underscored the importance of deploying these tools responsibly, especially as AI systems are increasingly applied in critical areas such as regulatory submissions and scientific research. He highlighted the potential of AI to serve as a valuable advisor, augmenting human expertise and improving decision-making processes in complex domains.

(Sidenote: we have plenty of our own thoughts on how to reliably ground AI for molecular modeling in truth which you can check out here.)

Mark Taisey on automation at Amgen

Mark Taisey, Senior Vice President of Global Regulatory Affairs and Strategy at Amgen, talked through automation in the company’s approach to regulatory submissions and post-approval changes, some recent key achievements include: 

• Accelerating BLAs: Historically, Amgen needed about 26 weeks from database lock to filing a Biologics License Application (BLA). Automation has significantly shortened this timeframe, including cutting the report-generation phase from 6 weeks to just 2-3 days. Today, the full BLA process takes 12 weeks, with ongoing efforts to reduce this further to 8 weeks by mid-next year.
• Post-Approval CMC Changes: Securing global approvals for post-approval Chemistry, Manufacturing, and Controls (CMC) changes once required up to 5 years—a delay that hindered the availability of improved therapies. By leveraging cloud-based platforms like Cumulus, Amgen is now enabling simultaneous submissions to regulatory agencies worldwide. The company aims to reduce this timeframe to less than 1 year, improving agility and addressing barriers to innovation in drug manufacturing.

Taisey’s remarks were in line with many through the event in growing collection of effective AI use in areas surrounding scientific research. But his remarks also highlight a key question for AI startups, given that it sounds like much of this AI work was done internally at Amgen: in what markets can startups provide a key advantage, versus pharma simply building it for themselves?

Liang Zhao's on AI in regulatory science

Liang Zhao, a professor at UCSF and former head of Quantitative Medicine at the FDA, provided perspective on the potential for AI to transform regulatory science. He emphasized that the FDA operates at the confluence of science, regulation, law, and public health—making it uniquely positioned to benefit from integrative technologies with many corollary applications. Areas of opportunity include:

• Streamlining Workflows: Zhao noted that AI has the capacity to simplify regulatory workflows, reducing time-consuming processes and enhancing overall efficiency. From drafting regulatory documents to analyzing complex datasets, AI could help bridge the gap between limited human resources and ever-growing regulatory demands.
• Knowledge Integration: Regulatory science generates vast amounts of data that often remain siloed, making it difficult to access and utilize effectively. Today’s AI is well positioned to consolidate and synthesize this information, enabling decision-makers to draw on a more comprehensive understanding of historical and real-time data. This could reduce redundancy, enhance consistency in decision-making, and ultimately improve the FDA’s ability to respond to emerging challenges.

Zhao’s take followed a slightly different tact from other speakers in the panel underlining the opportunity not just to speed up processes with AI but to utilize AI as a tool to reduce siloes between science, law, and public health initiatives. Zhao sees this helping regulatory bodies move beyond reactive approaches, enabling proactive strategies for assessing risks, optimizing resource allocation, and improving transparency.

Janet Woodcock on Knowledge Management

Janet Woodcock, former FDA Principal Deputy Commissioner, shed light on the persistent knowledge management challenges within the FDA and the opportunities for AI to address inefficiencies. (She also had some excellent zingers on the regulatory field’s habit of writing ‘endless deathless prose’). She reflected on how legacy processes, designed for a pre-digital era, have created bottlenecks in regulatory operations, with an overemphasis on exhaustive documentation rather than actionable insights.

Inside of this context, the largest opportunities for AI include: 

• Reducing Administrative Burden: namely by relieving the "busywork" that dominates regulatory workflows. For instance, AI could handle tasks like compiling documentation, cross-referencing historical data, and drafting routine reports. This would free up human experts to focus on areas requiring nuanced judgment, such as risk assessment and decision-making.
• Enhancing Structured Data Utilization: the ability of AI to accommodate minor inconsistencies, heterogeneous data structures,  or semi-structured data without losing functionality is a big win for the resiliency of data management automations in regulatory science. 

The broader implication of Woodcock’s take is that regulatory bodies are standing in front of an opportunity to make their knowledge ecosystems significantly more accessible and dynamic. This would not only streamline workflows but also enable faster responses to evolving challenges in public health and drug approval processes.

‍M&A Landscape in Drug Discovery and Biotech

JPM never fails as a source of deals and drama. We dipped into a panel on M&A in biotech to get a read from insiders. 

Centerpoint Partners painted a dynamic picture of the current and future landscape. Despite macroeconomic pressures, such as high yields and declining biotech indices (e.g., XBI down 1-2%), the fundamentals of innovation remain strong, driving significant activity in both late-stage and early-stage transactions.

Recent trends:

• Major Deals in Focus: The sector has seen prominent transactions such as Johnson & Johnson's $15 billion acquisition of Intracellular and Lilly's purchase of Scorpion. These deals reflect the appetite for both large-scale commercial transactions and earlier-stage acquisitions to bolster pipelines.
• Drug pipeline needs: Patent cliffs loom large for many pharmaceutical companies, prompting an urgent need to replenish pipelines. While the weight-loss drug market, led by players like Lilly and Novo Nordisk, is commanding attention, the broader search for transformative therapies continues. Notably, areas like neuroscience, cell therapy, and next-generation VEGF-PD1 bispecifics are gaining traction.
• China's Growing Influence: The rise of China as a biotech hub was underscored, with a significant portion of licensing deals now originating from Chinese firms—rising from negligible levels a decade ago to accounting for one-third of current deals. This shift reflects China’s growing role as both a competitor and collaborator in the global drug discovery ecosystem.

Market Context:

• Transaction Volume and Scale: In 2023, the industry witnessed 23 transactions totaling $140 billion. While 2024 showed slightly slower momentum with 18 $1B+ deals, the expectation for 2025 is a resurgence of larger-scale transactions as companies seek to maintain their competitive edge.
• Investor Sentiment: Despite macroeconomic challenges, biopharma remains a sector ripe for innovation. With high stakes for addressing unmet medical needs, M&A continues to serve as a critical avenue for growth and discovery.

The panel really focused on the importance of identifying differentiated assets in crowded or emerging markets. For example, in obesity—a field currently dominated by a few players—future acquisitions may hinge on innovative approaches or underutilized mechanisms. Similarly, neuroscience remains an area of great interest, with significant unmet needs and high revenue potential for successful therapies.

Let’s continue the convo!

We love seeing what the broader community is up to every JPM. While there are significant challenges to some AI applications and determining the outcome of recent political swings, we came away seeing real tangible momentum at many orgs dealing with AI adoption. Drug discovery is such a monumental undertaking, it’s humbling to see the challenges and innovations at each step in the process. If you chatted with us at JPM, or are just keeping tabs on what’s next in molecular simulation and AI in sciOps, feel free to reach out to see how we can partner (or give our new AI assistant Balto a try), we would love to chat!