Astellas needed to shorten the slow handoff between experimental scientists and computational chemists in early drug discovery so more researchers could use structure prediction and affinity screening without waiting on scarce specialist capacity.
Reported impact
Astellas cut the number of compounds needing experimental evaluation by more than 90 percent, shortened Hit-ID timelines by more than 70 percent, reduced complex-model response time from over 20 seconds to 7 seconds, and expanded the system to more than 2,000 total users within four months.
Astellas Pharma turned structure prediction and affinity screening into a broadly accessible internal service by deploying NVIDIA Boltz-2 NIM as a production microservice for experimental scientists. Instead of waiting for specialized chemistry handoffs, researchers submit biomolecular hypotheses into an application built around Boltz-2 NIM, with NVIDIA TensorRT speeding complex modeling and activity prediction. The service also draws on NVIDIA BioNeMo tools and DGX H100/H200 resources from Astellas' broader AI research environment. Predictions are used to prioritize compounds before wet-lab assays so only the most promising candidates move forward. That shifts early discovery from an expert bottleneck into a self-service loop that speeds hit identification, reduces experimental volume, and broadens access to molecular AI across the company.
Researchers submit target and compound hypotheses into an internal application built around Boltz-2 NIM. TensorRT-optimized predictions score structures and binding affinity, Astellas prioritizes the best candidates, and only the shortlisted compounds move into laboratory assays for hit identification.
AI-guided prioritization reduced the number of compounds requiring experimental evaluation to just 10, resulting in a 90% reduction in assays needed.
Applications incorporating Boltz-2 NIM have been provided company-wide; as of four months post-release, there are over 50 regular users, and the total number of users has exceeded 2,000.
The full process, from hypothesis generation to Hit-ID, was completed in less than one month.
Using Boltz-2 NIM optimized with TensorRT, results were obtained in just 7 seconds.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Sep 22, 2026 DELFI Diagnostics DELFI Diagnostics needed to speed genomic analysis for cancer screening and therapeutic monitoring, but CPU-bound alignment and somatic variant-calling steps created turnaround-time and cost bottlenecks that limited how… 10–20 business days DeepSomatic · MutectCaller · NVIDIA Parabricks NVIDIA Sample Apr 23, 2026 Biofy Hospitals needed pathogen identification and antibiotic-resistance analysis much faster than five-day culture and antibiogram workflows, but CPU-bound genomic processing and limited AI expertise made it hard to turn… 100,000+ synthetic DNA sequences generated by the biology model Abby · H100 Tensor Core GPUs · NVIDIA CUDA-accelerated libraries NVIDIA Sample Apr 8, 2026 SimBioSys Traditional breast-cancer presurgical planning relied on grayscale MRI images and short radiology summaries, which made tumor geometry hard to interpret, contributed to variable surgical decisions and unnecessary… above 20 percent MONAI · NVIDIA A100 Tensor Core GPUs · NVIDIA CUDA-X 