Astellas adopts Boltz-2 NIM to speed up drug discovery

Astellas Pharma had identified a bottleneck in drug discovery: computational chemists had to evaluate large numbers of compounds, slowing down the process. To address this, they integrated Boltz-2 NIM, an AI model that predicts both protein structure and compound binding affinity, into their workflow. The tool, optimized for NVIDIA GPUs, allowed medicinal chemists to run simple simulations themselves, reducing reliance on computational chemists for routine tasks. The company also developed a browser-based frontend and an application that converts structural formulas into SMILE notation, enhancing usability. Results were dramatic: in one project, the number of compounds for assay was cut to 10, reducing assay count by over 90% and time by over 70%, with all 10 hits confirmed. As of February 2026, the application had over 50 regular users and over 2,000 cumulative uses. Astellas plans to expand NIM deployment with additional models for other modalities, aiming to further democratize AI-driven drug discovery.