In a move that signals a seismic shift in pharmaceutical R&D, Schrödinger, the venerable scientific software firm founded in 1990, has announced a major strategic collaboration and software agreement with Bristol Myers Squibb (BMS). At the heart of this partnership is "Bunsen," Schrödinger’s newly unveiled agentic AI co-scientist. By acting as an intelligent execution engine for complex computational chemistry, Bunsen promises to automate and scale the drug discovery process to unprecedented levels, effectively moving the industry from manual computational oversight to autonomous, agent-led molecular exploration.
The Genesis of the Partnership
The relationship between Schrödinger and BMS is not a new development, but rather the maturation of a long-standing technical marriage. Pat Lorton, Schrödinger’s Chief Technology Officer and Chief Operating Officer, characterizes the history between the two companies as deeply collaborative. For years, BMS has utilized Schrödinger’s suite of software, including the widely adopted LiveDesign platform—a tool that enables medicinal and computational chemists to centralize data, visualize proposed molecular structures, and share in silico predictions in real-time.
The latest agreement, however, transcends mere software licensing. It is a developmental roadmap. BMS and Schrödinger have committed to co-developing new functionalities within Bunsen and enhancing RetroSynth, Schrödinger’s AI-driven synthesis planning platform. This evolution reflects a broader trend in Big Pharma: moving away from fragmented, tool-based workflows toward integrated, "agentic" environments where the software does not just provide data, but actively participates in the scientific decision-making process.
Chronology: From Wind Tunnels to Neural Networks
To understand the significance of Bunsen, one must view it as the latest iteration in a long technological evolution. Lorton draws a compelling analogy between modern drug discovery and the history of aerospace engineering.
In the mid-20th century, aircraft design was an empirical, labor-intensive process defined by the construction of physical wind tunnels and balsa wood models. It was the only way to measure drag and refine aerodynamics. Decades later, the introduction of computational fluid dynamics and supercomputing allowed engineers to simulate these conditions, drastically reducing the reliance on physical prototypes.
Pharma is currently undergoing a parallel transformation. In its early years, Schrödinger focused on physics-based simulations. The turning point arrived in the early 2010s with the proliferation of GPUs, which allowed for the large-scale implementation of Free Energy Perturbation (FEP) methods. By 2020, as evidenced by its IPO filing, the company was already perfecting "Large-Scale Molecule Exploration" and "Large-Scale Molecule Evaluation."
The final piece of the puzzle—the "agentic" layer—was catalyzed by the emergence of advanced coding agents like Claude Code in early 2025. These tools demonstrated that AI could not only write code but also test, debug, and iterate on its own output. Schrödinger realized that its existing suite of highly accurate, physics-based "expert tools" was ready to be orchestrated by a similar "intelligent agent" capable of interpreting quantitative results and deciding the next logical scientific step.
Supporting Data: Scaling the Funnel
The primary challenge in drug discovery has always been the "funnel" problem: the need to evaluate an astronomical number of potential compounds to find a handful of viable drug candidates.
Just five years ago, Schrödinger’s platform could enumerate roughly 100 billion compounds around a lead molecule in a weekly cycle. Today, that number has surged into the trillions. This expansion is not merely about volume; it is about selectivity. Because Schrödinger can now predict a wider array of molecular properties—such as potency, selectivity, and toxicity—simultaneously, the "top of the funnel" must be significantly wider to ensure that the 50 candidates emerging at the end are of the highest possible quality.
This push for massive scale coincides with a massive investment in hardware. BMS is currently deploying one of the most powerful supercomputers in the pharmaceutical industry, utilizing an NVIDIA DGX SuperPOD built on eight DGX Vera Rubin NVL72 systems. This follows similar moves by Lilly and Roche, who have both recently brought online massive Blackwell GPU-powered AI factories. These supercomputers provide the raw "horsepower" necessary to run the physics-based calculations and large-language model (LLM) workloads that drive modern discovery.
Official Responses and Strategic Intent
The adoption of Bunsen is a cornerstone of BMS’s "Predict First" strategy, which seeks to integrate AI-driven predictions into every stage of its small-molecule and large-molecule programs.
Stephen Johnson, Vice President of Computational Sciences at Bristol Myers Squibb, stated that Bunsen allows the company’s scientists to "think differently" about molecular design space. This is a critical distinction: the tool is designed to augment human intuition rather than replace it.
Erin Davis, BMS Vice President of Research Business Insights & Technology, emphasized the autonomy of their current approach, noting that the company is actively building its own foundational models to support large-scale molecular predictions. Schrödinger has positioned Bunsen to be "LLM-agnostic" to accommodate this. According to Lorton, while Bunsen performs exceptionally well with models like Claude, it is engineered to be compatible with open-weight models that pharmaceutical companies may choose to host locally on their own supercomputing infrastructure. This gives firms the flexibility to maintain control over their data and proprietary models while leveraging the orchestration power of the Bunsen engine.
Implications: Addressing the Talent Bottleneck
Perhaps the most significant implication of the Bunsen collaboration is its potential to alleviate the acute shortage of computational chemistry expertise.
Currently, the industry faces a structural bottleneck. While Big Pharma is investing billions in compute, there is a lack of human experts to manage the workflow. In many organizations, a single project may receive only a fraction of a computational modeler’s time, forcing them to spend precious hours on the "arduous task" of manual data assembly and workflow configuration.
Bunsen acts as a force multiplier. By providing a natural language interface, it allows a scientist to act as a project manager, directing the agent to "run this modeling campaign" and having the agent handle the technical minutiae. This democratization of expert-level workflows is, according to Lorton, the "equivalent of a postdoc" for the medicinal chemist.
The Role of Human Oversight
Despite the "agentic" moniker, Schrödinger is emphatic that Bunsen remains under human control. The platform offers a granular "human-in-the-loop" setting, where the agent proposes a strategy, waits for approval, and returns with questions when it encounters ambiguity.
Lorton stresses that the most effective use of these tools is not an "open-ended request," but a guided collaboration. An experienced chemist who understands a target’s unique hydrogen bonding or the specific pocket of a protein can direct the agent to focus on high-yield areas. In this sense, the AI does not replace the chemist; it offloads the "heavy lifting" of simulation and enumeration, allowing the scientist to focus on the high-level strategy and hypothesis generation that defines truly innovative medicine.
The Future of Drug Discovery
As the pharmaceutical industry continues to pour capital into GPU-driven supercomputing, the divide between companies that merely use AI and those that integrate it into an agentic workflow will only grow. By partnering with BMS to deploy Bunsen, Schrödinger is betting that the future of drug discovery belongs to those who can effectively orchestrate a symphony of predictive tools, physics-based simulations, and autonomous agents.
The journey from the "balsa wood" era of drug design to the age of the "AI co-scientist" is nearly complete. With trillions of molecules now within the reach of computational exploration, the challenge is no longer about finding enough candidates—it is about having the intelligence to navigate the vast, digital landscape of chemistry to find the ones that will save lives. Through the Bunsen-BMS collaboration, the industry is taking a definitive step toward making that vision a reality.
About the Author
