---
title: "Run NVIDIA BioNeMo NIM Microservices for Protein Structure Prediction in Claude Science"
description: "Claude Science is used with the NVIDIA BioNeMo Agent Toolkit and NVIDIA NIM microservices to investigate how the predicted Seh1 structure changes alone versus with proposed partner C1HCX1. The workflow retrieves UniProt sequences, generates unpaired and species-paired multiple-sequence alignments with MSA Search, and sends those inputs to OpenFold3 and Boltz-2 for independent monomer and heteromer predictions. Using MSA input, heteromer iPTM reached 0.85 for OpenFold3 and 0.82 for Boltz-2, versus 0.14 and 0.19 without it; five-sample runs had standard deviations no greater than 0.006. Both models placed partner β-strands at Seh1’s WD40 propeller closure, while core Cα-RMSD remained 0.68 Å and 0.65 Å, respectively, but the predicted interaction remains experimentally unverified."
---

# Run NVIDIA BioNeMo NIM Microservices for Protein Structure Prediction in Claude Science

[NVIDIA](https://yomu.fyi/company/nvidia-developer-blog) · Michelle Horton · Aug 31, 2026

**Type:** Tutorial

## Summary

Claude Science is used with the NVIDIA BioNeMo Agent Toolkit and NVIDIA NIM microservices to investigate how the predicted Seh1 structure changes alone versus with proposed partner C1HCX1. The workflow retrieves UniProt sequences, generates unpaired and species-paired multiple-sequence alignments with MSA Search, and sends those inputs to OpenFold3 and Boltz-2 for independent monomer and heteromer predictions. Using MSA input, heteromer iPTM reached 0.85 for OpenFold3 and 0.82 for Boltz-2, versus 0.14 and 0.19 without it; five-sample runs had standard deviations no greater than 0.006. Both models placed partner β-strands at Seh1’s WD40 propeller closure, while core Cα-RMSD remained 0.68 Å and 0.65 Å, respectively, but the predicted interaction remains experimentally unverified.

## Context

The workflow addresses how the predicted structure of the Seh1 protein C1GY11 differs when modeled alone versus with the proposed partner C1HCX1. It also addresses the challenge of selecting, formatting, and operating specialized protein-folding tools within an agentic scientific workflow.

## Approach / What changed

Claude Science uses BioNeMo Agent Toolkit skills to retrieve sequences, generate unpaired and species-paired MSAs with the MSA Search NIM, and run monomer and heteromer predictions through OpenFold3 and Boltz-2 NIMs. The workflow preserves requests, responses, alignments, structures, metrics, confidence fields, runtime details, and errors for review.

## Takeaways

- The GPU-based setup calls for an NVIDIA L40S or H100 and about 700 GB of storage, including roughly 490 GB for the UniRef30-only MSA Search database and 30–40 GB for the Boltz-2 and OpenFold3 containers.
- OpenFold3’s monomer confidence falls from pLDDT 82 to 36 without an alignment, while Boltz-2’s monomer score changes from 0.79 to 0.73; both models still require MSA input to place the interface confidently.
- Larger sampling budgets did not replace evolutionary input: with-MSA and no-MSA interface results stayed effectively unchanged, while the two model families agreed within 0.03 iPTM when given the alignment.

**Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [GPU](https://yomu.fyi/topic/gpu), [Model Serving](https://yomu.fyi/topic/model-serving)

- Source: [NVIDIA](https://developer.nvidia.com/blog/run-nvidia-bionemo-nim-microservices-for-protein-structure-prediction-in-claude-science)
- Source URL: https://developer.nvidia.com/blog/run-nvidia-bionemo-nim-microservices-for-protein-structure-prediction-in-claude-science
- Ingested by Yomu: 2026-09-01T00:00:55.861Z

[Read original post](https://developer.nvidia.com/blog/run-nvidia-bionemo-nim-microservices-for-protein-structure-prediction-in-claude-science)