PharmaOpinion

Astellas Cut Lab Tests by 90% in a Drug Screen Using Boltz-2

Boltz-2 helped Astellas narrow a drug screen to 10 compounds, all of which showed the desired activity in lab tests. The strongest performed comparably in the lab to a benchmark compound already in clinical trials.

Answer
Astellas Cut Lab Tests by 90% in a Drug Screen Using Boltz-2

The brief

  1. Astellas Pharma reduced lab assays by 90% in a drug screening project using Boltz-2, a predictive model developed by MIT and Recursion.
  2. All 10 compounds tested after predictions with Boltz-2 showed the desired activity, with the strongest hit comparable to a benchmark compound in clinical trials.
  3. Astellas created a browser app allowing medicinal chemists to run routine predictions independently, streamlining the decision-making process.

Deployment

Company
Astellas Pharma
System
Boltz-2
Stage
Live
fewer lab tests
90%

Astellas Pharma’s researchers had a membrane protein they wanted to target and a library of compounds they could test against it. Each compound sent to the lab would take an assay. The team first ran predictions with Boltz-2, then picked 10 compounds to test. All 10 showed the activity the researchers were looking for, according to an NVIDIA case study.

The brief

  1. Astellas Pharma reduced lab assays by 90% in a drug screening project using Boltz-2, a predictive model developed by MIT and Recursion.
  2. All 10 compounds tested after predictions with Boltz-2 showed the desired activity, with the strongest hit comparable to a benchmark compound in clinical trials.
  3. Astellas created a browser app allowing medicinal chemists to run routine predictions independently, streamlining the decision-making process.

Deployment

Company
Astellas Pharma
System
Boltz-2
Stage
Live
fewer lab tests
90%

NVIDIA reported about 90% fewer assays on that project. The strongest hit showed activity in the lab comparable to a benchmark compound already in clinical trials. The case study gives no test results for the compounds Astellas screened out, leaving open whether the model passed over other useful hits.

A Simulation Chemists Can Run Themselves

“Until now, simulations were performed exclusively by computational chemists,” said Kazuya Nagaoka, an Associate Director in Astellas Pharma’s DigitalX research and development team. Medicinal chemists design compounds and decide what to make next. Astellas built a browser app so they could run routine predictions as they worked through those decisions.

Boltz-2, developed by researchers at the Massachusetts Institute of Technology and Recursion, predicts how a protein and a compound fit together in three dimensions and estimates how strongly they bind. Astellas runs NVIDIA’s packaged version, Boltz-2 NIM, on Tokyo-1, a shared computing platform for drug discovery in Japan. A separate application developed by Astellas lets chemists draw compounds and queue predictions for several protein-compound combinations.

Keigo Ide, Digital Biological Modality Discovery Team Lead at Astellas, said “results were obtained in just 7 seconds” for one workload using the version optimized with NVIDIA TensorRT. The open-source Python implementation took more than 20 seconds on that workload. For a chemist weighing several compounds, the app also removes the handoff for a routine run.

How Far the Workflow Can Travel

By the end of February 2026, the browser app had reached more than 2,000 users, with more than 50 using it regularly, according to NVIDIA. Astellas has also put Boltz-2 into a protein and peptide design workflow. RFdiffusion proposes structures and ProteinMPNN proposes sequences to fit them, then Boltz-2 predicts how the designs might bind.

The membrane-protein project gives Astellas a result to compare against future screens. Across more targets, the measure to watch is how many assays it takes to find a hit worth pursuing. For Yosuke Matsumura, a Senior Researcher at Astellas, complex simulations still go to specialists. Running the routine ones himself means he can “apply the results to our next ideas.”

Written by Amit Raja Naik

Join the discussion

0 answers

People who run these systems answer here. Your verified role is shown, never your employer unless you choose.

Sign in with LinkedIn to answer

Read next

MachineCon USA · from AIM Media House

Where enterprise AI is decided

An invitation-led day for the CIOs, CDOs and Chief AI Officers who own enterprise AI strategy.

  • New York5th edition · 583 Park Avenue
  • Dallas1st edition

Register for MachineCon USASee MachineCon USA →

  • Keynote on the main stage, MachineCon New York 2025
  • AI150 award winners on stage, MachineCon New York 2025
  • At the tables, MachineCon New York 2025
  • Networking, MachineCon New York 2025