How do I add filtering between stages?

Add a small "filter" symbol (a funnel icon) on each inter-stage arrow with a one-line criterion (e.g., "pLDDT > 80, Ramachandran clean").

PromptsPipeline

PipelineText → ImageMid creditEN

AI Protein Design Workflow

Backbone generation, sequence painting, structure prediction and binding-energy calculation.

When to use this prompt

For computational structural biology and protein engineering papers.

The prompt

An AI protein design workflow, four stages flowing left-to-right.

Stage 1 — Backbone Generation
- A diffusion-style model (RFdiffusion-class) generates a 3D protein backbone scaffold from random noise, conditioned on a binding hotspot specification.
- Show: random noise -> ribbon backbone schematic.

Stage 2 — Sequence Design
- A sequence-design model (ProteinMPNN-class) paints amino acid identities onto the fixed backbone.
- Show: backbone -> backbone + colored side-chain residues.

Stage 3 — Structure Prediction
- The designed sequence is folded with an AlphaFold-class model in complex with the target.
- Show: predicted complex with a confidence heatmap (pLDDT) overlaid in blue-to-red gradient.

Stage 4 — Binding Energy Estimation
- A scoring function (PRODIGY-class) computes binding free energy from the predicted complex.
- Show: numeric output ΔG = -X.X kcal/mol.

Style: scientific infographic, white background, restrained biomedical palette (navy, teal, coral), thin connecting arrows, sans-serif labels. Suitable for Nature Methods or computational biology journals.

Variations

With wet-lab validation arm

Append a Stage 5 "Wet-Lab Validation" showing recombinant expression, biolayer interferometry binding measurement, and a comparison plot of computed vs measured K_d.

Tips

Mention each model class by analogy (RFdiffusion-class, AlphaFold-class) — generators reproduce the structure even if the exact name varies.
Show the pLDDT heatmap explicitly — it is the visual signal most readers expect at the structure step.
End with a numeric output. Workflows that end on a vague step lose impact.