NVIDIA BioNeMo Agent Toolkit: Accelerating AI for Life Sciences Researchers with Claude Science

The life sciences sector has embraced computational scale, marking a significant transition. For over a decade, NVIDIA has developed a comprehensive GPU-accelerated computing architecture featuring hardware, frameworks, libraries, models, microservices, and specialized tools designed to enable researchers to execute advanced workflows and accelerate iteration.

This week, Antropic introduced Claude Science, an innovative AI workbench tailored for scientific research, empowering scientists to engage with AI agents through natural language for comprehensive task completion.

Claude Science seamlessly integrates with the NVIDIA BioNeMo Agent Toolkit, providing researchers with efficient access to tools within their workflows. This toolkit encapsulates NVIDIA’s acceleration capabilities as callable skills, enabling Claude Science to choose suitable tools, generate valid inputs, and execute workflows efficiently. Connect all of this to NVIDIA compute resources deployed anywhere, ensuring that NVIDIA’s accelerated models, libraries, and NIM microservices are integrated into the research environment.

Leading pharmaceutical companies are leveraging NVIDIA technology to enhance AI-driven research across drug discovery, genomics, medical imaging, molecular design, and protein engineering, with 18 out of the top 20 pharmaceutical companies utilizing this technology. Explore NVIDIA BioNeMo to understand its extensive role across the ecosystem.

Leading the AI-Driven Era of Scientific Discovery

Claude Science empowers scientists to translate their research inquiries into actionable insights without requiring manual model configuration, endpoint setup, or software environment adjustments. With access to the NVIDIA BioNeMo Agent Toolkit, scientists can take advantage of accelerated workflows and models like Evo 2, Boltz-2, and OpenFold3, ensuring faster analyses that benefit from computational acceleration.

Researchers initiate their tasks by articulating research objectives in natural language, such as analyzing genomic sequences, predicting protein structures, and designing potential binding agents. Claude Science interprets these requests and orchestrates workflows through preconfigured, domain-specific agents that are adept at managing established workflows in genomics, proteomics, single-cell analysis, chemoinformatics, and clinical research.

The BioNeMo Agent Toolkit equips these agents with vital context, linking each step to the most relevant NVIDIA scientific capabilities. Each skill comes with information regarding its purpose and required inputs, aiding agents in preparing and executing workflows and returning outputs for evaluation.

This collaborative process creates an iterative loop between scientific reasoning and enhanced computational work, allowing scientists to review outputs, refine their inquiries, and determine subsequent steps while focusing on the core scientific questions.

A notable example includes generating improved inhibitors for common cancer targets. In this workflow, a researcher begins with a known cancer-related antigen variant and requests Claude to propose several potential inhibitors. Claude Science, combined with the BioNeMo Agent Toolkit and NVIDIA NIM microservices, accelerates high-throughput prediction, optimization, and validation of inhibitors.

A Scientifically Robust Foundation for Agents

AI agents are designed to reason, plan, and utilize tools to accomplish tasks. In the realm of life sciences, these tools are often specialized computational workflows.

Autonomous AI scientist agents do not operate in isolation. They might need to fingerprint a chemical library, cluster promising discoveries, generate top candidate conformers, analyze genomic contexts, and compare perturbation responses before suggesting the next experimental step.

The efficiency of each step depends on scientific tools; hence, the agent’s speed is contingent on the performance of those tools.

The NVIDIA BioNeMo Agent Toolkit bestows scientific agents with rapid tools essential for thriving in the fast-paced scientific landscape. Key components include:

• NVIDIA Parabricks accelerates genomic analysis from hours to mere minutes, allowing agents to incorporate genomic context into decision-making almost in real-time.
• RAPIDS-Single Cell, developed by scverse, reduces a 1.3 million cell preprocessing and clustering workflow from 52 minutes to just 25 seconds, integrating single-cell analysis rapidly into the inference cycle.
• nvMolKit, which accelerates chemoinformatics operations such as similarity searches and conformer generation by up to 3,000 times, enabling agents to generate results almost instantaneously while exploring large chemical spaces.
• NVIDIA BioNeMo open models that provide essential biomolecular functionalities accelerated by NVIDIA libraries, ensuring dedicated models for every stage of the workflow.
• BioNeMo NIM microservices enable packaging of models as enterprise-ready inference endpoints (containerized microservices) with a complete accelerated software stack pre-integrated and finely tuned for high-performance inference, allowing agents to utilize a stable API for production deployment.

The NVIDIA BioNeMo Agent Toolkit is open-source and hardware-agnostic, allowing researchers to employ the same scientific skills across various agent frameworks and research platforms. The toolkit and its features are currently accessible through NVIDIA developer resources and on GitHub.

Researchers can access BioNeMo-powered workflows through Anthropic’s Claude Science, which officially launches its public beta today. During this phase, Anthropic invites feedback from researchers regarding the integration and additional domain expertise required.