protenix-structure-prediction

name	protenix-structure-prediction
description	Run pinned Protenix structure prediction for protein, nucleic-acid, ligand, antibody-antigen, template, MSA, or constrained complexes. Use Protenix-v2 or another explicitly selected released model with a recorded training-data cutoff and inference budget.
license	MIT

Protenix Structure Prediction

Use $cx-modeling-problem-execution for concrete inputs and continue through execution, comparison, and review.

Gate and workflow

1. Ask once before installing a pinned Protenix release, downloading weights or databases, using public MSA services, and running GPU/remote compute.
2. Validate the JSON entities, ligand states, templates, constraints, MSAs, and RNA features. Pin model name/checksum, code version, training-data cutoff, samples/seeds, inference-time scaling budget, kernels, device, and precision.
3. Smoke-test the official example. Run protenix pred into artifacts/<run-id>/protenix/; retain inputs, features, models/configs, structures, confidence, logs, timings, and every candidate/failure.
4. Report performance as specific to the selected model/cutoff. Inspect sample diversity, interface/ligand plausibility, clashes, constraints, MSA/template dependence, and uncertainty; compare another predictor for strong claims.
5. Record with $science-provenance; review with $science-review.

Boundaries

• More inference samples increase selection opportunity and compute; report the full budget and avoid comparing against a lower-budget baseline unfairly.
• Structure confidence is not experimental interaction or affinity evidence.