Protein structure prediction with Claude Code & Claude Cowork

Overview

Structure prediction used to be a paper-sized milestone. Now it is a Tuesday afternoon. This course teaches you how to take a protein sequence — one you picked from your own genome data, a pathogen surveillance set, or a literature target — predict its structure, analyze it, and generate testable functional hypotheses. Claude Code runs the prediction and analysis pipelines. Claude Cowork is where your specialist agents argue about what the structure means.

How Claude Code and Claude Cowork fit

• Claude Code runs locally or on a GPU box. It manages sequence inputs, calls AlphaFold-class prediction tools, runs structural comparisons, and keeps every intermediate artifact versioned in git.
• Claude Cowork hosts a multi-agent workspace: a Structural Biologist agent (reads pLDDT, PAE, and packing), a Functional Annotator agent (maps domains, binding sites, orthologs), and a Literature Reviewer agent (cross-checks claims against published work). You act as the decision-maker.

Who it's for

• Biologists, chemists, and engineers moving into computational structural biology
• Graduates of the personal genome bioinformatics course who want to interpret their variants at the protein level
• Drug discovery and antibody engineering-adjacent roles exploring AI structural tools

What you'll build

• A reproducible sequence-to-structure pipeline orchestrated by Claude Code
• A Claude Cowork project with Structural, Functional, and Literature agents collaborating on each target
• A one-page "structural brief" per protein: predicted fold, confidence, likely function, suggested experiments

Prerequisites

• Comfort with the command line and Python environments
• Access to a GPU machine or cloud credits sufficient for AlphaFold-class inference
• Basic biology literacy (domains, binding sites, conservation)

Tools and setup

1. Pick a set of target sequences with clear scientific interest
2. Install a prediction toolchain (AlphaFold, ESMFold, or equivalent) driven by Claude Code
3. Create a Claude Cowork project and invite the three specialist agents

Modules

Module 1: Sequence to structure with Claude Code

You will build a pipeline that pulls sequences, runs predictions, scores confidence, and stores structures and metrics in a versioned project.

Module 2: Multi-agent interpretation in Claude Cowork

You will walk through a predicted structure with your Cowork team, letting Structural, Functional, and Literature agents cross-reference each other's claims before writing anything down.

Module 3: Turn structures into experiments

You will translate structural insights into specific, testable wet-lab or computational follow-ups — mutations, binding assays, or comparative analyses.

Deliverable

A structural-biology project folder with reproducible predictions, a Cowork discussion transcript per target, and a structural brief that recommends concrete next experiments.

Common mistakes

• Trusting high-confidence predictions on sequences outside the training distribution
• Ignoring pLDDT and PAE and treating every model as definitive
• Writing conclusions before the Literature Reviewer agent has checked the claim

Next steps

Move into protein-protein interaction prediction, ligand docking, or AI-assisted protein design for your CRISPR or lab automation projects.