Anthropic Claude Science and NVIDIA BioNeMo Accelerate Scientific Research Through Automated Computational Workflows and Genomic Analysis
Natural language processing breakthrough goals look achievable with the recent launch of new research workbenches. Anthropic released Claude Science, a platform designed to handle complex scientific workflows automatically. As stated in the product announcement, the platform is integrated directly into the NVIDIA BioNeMo Agent Toolkit. This integration enables scientists to automate molecular biology, genomics, and chemistry computational workflows with a conversational UI directly from remote NVIDIA graphics processors.
For over ten years, top scientific equipment makers have built dedicated hardware and software libraries for the pharmaceutical industry. Such investments are already being adopted widely as 18 of the top 20 global pharma companies now use the BioNeMo platform in their research. With this new integration, those same tools are brought directly into the research environment rather than relying on custom servers, APIs, or software containers.
Using Claude Science, a researcher can write out a target goal in simple language, such as model a structure for a protein or look for gene sequences matching the criteria. The engine uses preloaded specialized agents to translate the goal and match it with the existing tool. The BioNeMo Agent Toolkit supplies the agents with the programmatic context required to send in data, run the right models, and send the result directly to the interface for review.
This automated loop affords the researcher instant access to advanced models such as Evo 2, Boltz 2, and OpenFold3, which can speed research greatly. For example, in oncology, a scientist might want a specific anticancer causing genetic mutation or antigen designed. They can ask the system to generate and optimize multiple inhibitor molecules while preverifying their practical utility in the laboratory. The agent then communicates remotely from various microservices to predict and develop molecules for molecular binding in mere hours instead of weeks or months.
Autonomous research agents require speedy and reactive computing tools so as not to bottleneck multi step processes. The BioNeMo Agent Toolkit packages a host of optimized scientific libraries such as Parabricks for genomic analysis. This tool creates 11 times the processing speed of other hardware and brings processing times from hours down to minutes, allowing an AI agent to actively leverage real time genomic data as it works through training.
Similarly, to speed cellular work, the toolkit employs RAPIDS single cell, a library created with scverse that cuts down the time to precluster a 1,300,000 cell data set from 52 minutes to a mere 25 seconds. Another tool called nvMolKit accelerates chemistry operations like molecular similarity searches and conformer creation by 3000x over the best hardware solutions at the speed of thought.
To make these tools ready for corporate pharmaceutical pipelines, the system uses BioNeMo NIM microservices, a containerized and preconfigured package launched for high speed inference. The microservices create consistent endpoints, allowing research agents to reliably call from their remote compute clusters. Lastly, because the BioNeMo Agent Toolkit is open source and demand driven, developers may obtain the code through GitHub and plug these scientific capabilities into their other custom agent frameworks.
