Highlights from the 2025 AWS Life Sciences Symposium’s Drug Discovery track

On May 6, more than 1,000 life sciences leaders from over 400 organizations gathered in New York City for the seventh annual Amazon Web Services (AWS) Life Sciences Symposium. Centered on the theme “Building for Breakthroughs: AI-powered Innovations Transforming the Pharmaceutical Value Chain,” the event featured 26 sessions led by 39 experts—offering an unfiltered look at what’s working (and what’s not) in the pursuit of AI-driven success in life sciences.

One message rang loud and clear: generative AI is no longer a future promise—it’s a present-day force reshaping how medicines are discovered, developed, and delivered. That real-world impact took center stage in the Drug Discovery Breakout Track, where industry leaders showcased how AI is accelerating research, enhancing scientific precision, cutting costs, and ultimately helping bring therapies to patients faster.

Breaking the Mold: The Case for Radical Rethink in Drug Discovery

We’re living in a golden age of biomedical innovation—where science, data, and technology are converging to unlock entirely new therapeutic frontiers. Yet drug discovery remains slow, costly, and uncertain. Nearly 90% of drug candidates fail in clinical trials, and bringing a single treatment to market can take over a decade. Meanwhile, despite billions invested, many diseases still lack effective therapies.

At the core of the challenge is biology’s staggering complexity—spanning thousands of cell types, over 20,000 genes, and countless molecular interactions, each with variable responses across individuals. Researchers face an almost infinite landscape of therapeutic possibilities, including 10⁶⁰ drug-like small molecules and as many as 20³² potential therapeutic antibodies. Meanwhile, the critical data needed to fuel discovery remains scattered, unstructured, and siloed across journals, databases, and proprietary systems. Traditional R&D tools—while powerful—weren’t built to handle this scale, complexity, or speed.

The challenge is economic as well. With an estimated $250 billion in revenue at risk from upcoming Loss of Exclusivity (LOE) over the next few years, pharmaceutical companies face a growing innovation gap. Acquiring external assets may offer short-term relief, but long-term success depends on internal reinvention.

To meet this moment, pharma needs a new R&D playbook—one that puts AI at the core. Not just to take more shots on goal, but to make each one faster, more accurate, and more likely to succeed.

This shift isn’t incremental—it’s transformational. And as speakers across the symposium made clear, that transformation is already underway.

Lab in a Loop: Genentech’s Vision for AI-first Research

Rich Bonneau, VP of Machine Learning for Drug Discovery at Genentech, shared how the company is reimagining early-stage research with AI at the center.

At the heart of this approach is the concept of a “lab in a loop”—a tightly integrated, iterative cycle where AI models, trained on experimental and clinical data, generate predictions that guide lab experiments. As new data is produced, it feeds back into the models to refine them and improve their accuracy. This continuous feedback loop allows researchers to improve predictions, explore vast chemical spaces, and simultaneously optimize over multiple therapeutic properties—dramatically accelerating discovery.

These advances are driven by vast, high-quality datasets built from decades of laboratory and clinical research at Genentech—fueling AI models to take on challenges once considered intractable. And the company is going further, integrating generative AI directly into its research workflows.

In collaboration with AWS, Genentech developed the gRED Research Agent—an intelligent assistant powered by Anthropic’s Claude Sonnet 3.5 running on Amazon Bedrock Agents. This tool automates labor-intensive tasks such as sifting through scientific literature and structured databases, freeing scientists to concentrate on complex, high-impact research.

The impact is profound. Processes that once took weeks now take minutes—in biomarker validation alone, Genentech anticipates saving over 43,000 hours of manual effort.

But it’s not just faster science—it’s science at a fundamentally new scale.

Watch the session recording | Access the presentation here

AbbVie’s holistic approach to Designing Molecules with BioFMs

The conversation naturally turned to another critical challenge in drug discovery: small molecule design. Jennifer Van Camp, Vice President of R&D Data Analytics at AbbVie, and Abhishek Pandey, Global Lead and Principal Machine Learning Scientist, shared how their team is tackling this problem with a bold, biology-first approach—leveraging biological foundation models (BioFMs) to drive innovation.

In a major computational milestone, AbbVie calculated druggability scores for the entire human genome—predicting which proteins could be effectively targeted by small molecules. This genome-wide assessment represents a step-change from traditional models that focus only on ligand properties. Instead, AbbVie’s approach leverages biological traditional models that focusing solely on ligand properties, and takes a more holistic approach of incorporating generalizable patterns of protein and ligand properties as well as their interactions through bioFM models like ESM-2 by EvolutionaryScale.

To turn this biological complexity into predictive power, AbbVie integrated ESM embeddings with ligand data and applied AffinityNet—a deep learning model that uses graph attention networks to layer protein, ligand, and interaction properties—to predict binding affinity with exceptional accuracy. The result: a more biologically informed, data-driven approach to small molecule discovery.

Powered by AWS infrastructure, scalable compute, and access to BioFMs through Amazon Bedrock, AbbVie can now turn vast amounts of biological data into actionable insights—faster than ever before.

Watch the session recording | Access the presentation here

Agentic AI: Reimagining the researcher’s role

Biotech startups Owkin and Bioptimus showcased how they’re pushing the boundaries of AI-driven scientific collaboration to advance our understanding of biology.

Jean-Philippe Vert, Chief R&D Officer at Owkin, and Thea Backlar, VP of Product, introduced the K Navigator—an AI-powered co-pilot designed for biologists, by biologists. Built on agentic systems and connected to vast datasets—including MOSAIC Window, the world’s largest spatial omics dataset in oncology—K Navigator enables researchers to explore, analyze, and visualize multimodal data using natural language prompts. It amplifies scientific intuition, accelerates workflows, and transforms complex questions into actionable insights—advancing target identification, cohort characterization, drug repositioning, and more.

Scientists can now access over 35 million scientific papers, design clinical trials, and interrogate patient data—all within one interface. But K Navigator is more than a productivity tool; it represents a new paradigm in research—where humans and AI work side by side to accelerate insights and explore scientific frontiers once out of reach. Already available to academics, Owkin envisions this platform as a stepping stone to Biological Artificial Superintelligence (BASI)—a collaborative system where AI augments human discovery at scale.

Meanwhile, French startup Bioptimus is harnessing biology’s inherent multimodality. At the symposium, Co-founder and Principal Research Scientist Zelda Mariet introduced H-optimus-1—a foundation model trained on millions of whole-slide pathology images. The model achieves best-in-class performance across key diagnostic and research applications—such as cancer subtype classification, tumor grading, biomarker detection, and digital slide quality control—by learning relationships across biological scales to drive deeper insights.

Available now on AWS Marketplace, healthcare and life sciences organizations can quickly deploy Bioptimus’s cutting-edge pathology model within their secure cloud environments and seamlessly integrate it into AWS-based workflows—accelerating time to insight and boosting efficiency in pathology and biomedical R&D.

And this is just the beginning. Bioptimus is currently developing M-optimus, a next-generation multimodal model that combines genomics, molecular data, imaging, and clinical records—designed to learn from life in all its complexity. Their ambitious goal: to create the world’s first universal AI foundation model for biology.

Watch the session recording | Access the presentation here

Building the Lab of the Future

Raveen Sharma, Managing Director of Healthcare and Life Sciences at Deloitte, shared a bold vision for the “Lab of the Future”—a digitally integrated, AI-powered ecosystem where computation and experimentation work in seamless harmony – one that connects dry and wet labs in a continuous, intelligent loop, accelerating discovery through real-time data and automation.

In this model, dry labs train, fine-tune, and refine AI models and perform computational experiments, while wet labs test hypotheses and generate high-quality experimental data that feed back into the system. By making this data FAIR (Findable, Accessible, Interoperable, and Reusable), the loop becomes self-improving—driving both speed and scientific rigor.

Deloitte and AWS are bringing this vision to life with modular, cloud-native solutions that automate data ingestion, integration, and governance—transforming fragmented lab data into FAIR (Findable, Accessible, Interoperable, Reusable) assets. Deloitte’s Lab of the Future accelerator bridges the gap between physical lab instruments and digital infrastructure, leveraging proven AWS services like AWS DataSync, AWS IoT Greengrass, Amazon S3, and Amazon DataZone—all centrally managed through AWS Control Tower.

The impact is significant: scientists spend less time managing data and more time driving discovery—accelerating the development of higher-quality therapeutic candidates and getting them to patients faster.

Watch the session recording | Access the presentation here

Breaking Down Data Barriers: Secure Collaboration through federated learning

Science advances fastest when it advances together—and nowhere is that more true than in drug discovery.

One of the biggest bottlenecks in advancing AI for modern drug discovery is access to high-quality protein and ligand structure data—essential for training effective models. While public databases are available, they lack the depth and diversity of structural conformations and molecular interactions needed the current demands in designing drugs for difficult targets. Biopharmaceutical companies have invested heavily in generating proprietary datasets that capture these interactions, but the confidential nature of this data has traditionally hindered collaboration and slowed individual progress.

Federated learning is helping to break down these barriers. In a compelling panel discussion, Justin Scheer, VP, In Silico Discovery, Johnson & Johnson, John Karanicolas, Head of Computational Drug Discovery, AbbVie, and Robin Röhm, CEO and Co-Founder, Apheris shared how the AI Structural Biology (AISB) consortium is leveraging federated learning to collaboratively train AI models across distributed datasets—without exposing the underlying data. This approach protects intellectual property while enabling shared insights that enhance drug specificity, improve molecular interactions, and accelerate the development of better therapies.

As models reach structure prediction accuracy on par with X-ray crystallography, federated learning is ushering in a new era of secure, collaborative drug discovery—enabling the industry to pool its collective intelligence and accelerate breakthroughs in precision medicine.

Watch the session recording

A New Era of Discovery

Ten years ago, few imagined AI would become integral to how we travel, shop, listen to music, and consume news.

Today, a similar shift is happening in drug discovery. AI is moving beyond proof-of-concept, driving real progress across critical R&D workflows. By helping scientists untangle disease biology and design better therapies faster, AI offers new hope for conditions that have long resisted treatment.

The 2025 AWS Life Sciences Symposium made one thing clear: AI is no longer just augmenting research—it’s transforming it, powered by foundation models, multimodal data, agentic AI, and cloud-native infrastructure. The future of drug discovery isn’t years away. It’s here—and it’s happening now.

At AWS, we’re proud to help our customers go further, faster. That’s why 9 of the top 10 global pharmaceutical companies globally trust AWS to power their generative AI and machine learning strategies.

Contact an AWS representative today to learn how we can help your organization accelerate what’s next.

AWS for Industries