dynotx / dynopsi

What is dynopsi?

Dynopsi enables inference and sampling for Dyno Psi-1, a Python package for de novo miniprotein binder design. Feed it target structures from RCSB PDB IDs, AlphaFold DB UniProt IDs, or local PDB/CIF files via YAML configs or Python scripts, specify crops, hotspots, and binder lengths/positions, and it generates backbone coordinates in PDB, CIF, or NPZ formats. Pair outputs with ProteinMPNN for sequences and AF2 for filtering—synopsis of a full binder design workflow without custom code.

Why is it gaining traction?

Modular pipelines let you matrix hotspots, crops, and binder centers for batch designs, with CLI commands like `dynopsi check` to validate featurization via colored PDB previews before GPU-heavy sampling. Supports ODE/SDE solvers with tunable steps, and repository GitHub Actions handle builds—streamlining what used to need brittle scripts. The metadata repository enables multiple perspectives on data like binder positioning, standing out from rigid design tools.

Who should use this?

Comp bio researchers and protein engineers designing miniprotein binders against drug targets or antigens. Perfect for teams using VSCode with repository GitHub tokens for private repos, iterating designs via repository GitHub Pages demos or API integrations, avoiding "repository GitHub not found" headaches on public structures.

Verdict

Grab it if miniprotein binders are your focus—CLI and YAML make prototyping fast despite needing CUDA GPUs. With 10 stars and 1.0% credibility, it's immature; docs shine but expect tweaks for production. Apache-2.0 licensed, worth forking for custom flows.

(198 words)