Accelerating Drug Discovery with Agentic AI and Computational Science

By Louis Culot, corporate vice president and general manager, Cadence Molecular Sciences (OpenEye)

We recently discussed the rise of the autonomous engineer—the transition from human-driven to human-supervised workflows in semiconductor design made possible with AI-driven orchestration and enabled by AI super agents. An important question for life sciences is whether a similar transformation may emerge for pharmaceutical research and drug discovery. While the concept of a fully autonomous scientist remains aspirational, we believe the industry is approaching an important inflection point as agentic AI systems begin to improve the accessibility, efficiency, and scalability of computational scientific workflows. To accelerate this transition, we are excited about extending our Cadence Molecular Sciences (OpenEye) collaboration with NVIDIA and incorporating the new NVIDIA BioNeMo Agent Toolkit into our Orion Agentic AI roadmap.

By orchestrating AI agents with domain-specific capabilities across biology, chemistry, and genomics, agent toolkits can help transform complex computational workflows into repeatable, agent-executable processes—from model selection and input preparation to execution, output evaluation, and recommendation of subsequent actions. We believe this type of orchestration will accelerate scientific iteration and broaden access to advanced computational methodologies.

At Cadence, we view our AI platform as a three-layer cake framework with accelerated compute and data as the base layer, principled simulation and optimization as the critical middle layer, and AI as the top layer. We believe the greatest value comes from the tight coupling of the top two layers, with AI and principled simulation reinforcing one another iteratively to deliver more optimized and physics-accurate results.

In semiconductor design, this architecture enables Cadence's specialized super agents—including the ChipStack AI Super Agent, ViraStack AI Super Agent, and InnoStack AI Super Agent—to operate under human supervision, orchestrating complex engineering tasks across system and chip development, and delivering substantial improvements in productivity and time to market. We believe a similar architectural model applies to molecular sciences, adapted appropriately for the scientific and experimental realities of biology.

Earlier this year, Cadence released substantial capabilities allowing pharmaceutical customers to make use of coding agents, including Claude Code and Codex, to develop drug discovery applications using Cadence's chemistry and protein biology toolkits. These industry-standard toolkits enable molecular structure representation and are used broadly in drug discovery informatics, virtual screening, hit identification, and predictive ML modeling. But like much of computer-aided drug design (CADD), these applications must be written by expert scientist/programmers, so they are still somewhat removed from the day-to-day workflows of end-user scientists.

Agentic AI enables us to take a big step forward, orchestrating tools (skills) not just from Cadence, but allowing us to integrate skills from the broader ecosystem to provide customers with a comprehensive flow incorporating leading scientific tools, models, and workflows. This is why Cadence is excited to extend our collaboration with NVIDIA to incorporate the BioNeMo Agent Toolkit into our agentic flow.

In a computational protein drug target exploration workflow, multiple specialized steps and tools are required. Protein structures must first be prepared for downstream analysis, followed by the setup and execution of molecular dynamics simulations, binding-site characterization, and interpretation of simulation outputs in a scientifically interpretable form. These stages are often iterative, requiring repeated refinement as new insights emerge. In this context, the Cadence Orion Molecular Design Platform can orchestrate workflows, leveraging the BioNeMo Agent Toolkit alongside Cadence scientific engines to run computationally rigorous methods for protein preparation, molecular dynamics, binding-pocket analysis, virtual screening, and generative chemistry. Rather than requiring researchers to manually coordinate fragmented computational tools, agentic systems can increasingly reason from scientific questions, select appropriate computational methods, evaluate results, and recommend subsequent actions.

Will fully autonomous scientific discovery emerge in the near term? In our view, the more immediate opportunity lies in intelligent scientific collaboration, with agentic systems increasingly functioning as computational co-scientists helping researchers navigate complex workflows, accelerate iteration, and shorten the path from scientific question to actionable insight.

Cadence's Orion Agentic AI solutions incorporating BioNeMo Agentic Tools are expected to be available to early access customers in the second half of 2026.

Contact us to learn more about Cadence Molecular Sciences.