How-to guides

Task-oriented instructions for running RFD4-Proteina: the config overrides and CLI flags for a specific job, not why the underlying mechanism behaves the way it does. Each page links back to its concept page for that background. Start at Quickstart if you haven’t run the model at all yet.

Preparing inputs Building an annotated CIF: the agent, YAML, and Python API routes.

Agent YAML/JSON spec Python API

Running the model The core generation command: sampling workflows, inference noise, TSR, and troubleshooting a run.

Multi-job runs FoldSeek clustering

Partial diffusion and inverse folding Starting a run from a structure you already have instead of from noise. Metric guidance Configuring the smoke test, metric and output keys, schedules, and a real guidance objective. Inference backends Running the workflow through the native model or BioIR, and what each backend supports. Reward guided search Configuring beam search checkpoints, rewards, continuation criteria, and novelty.

Evaluate with metrics Turning on metrics inline or offline, running refolding, and aggregating results across jobs.

Custom metrics Offline metrics

Refolding Turning on the AF3/OF3 oracle, what it costs, and running a design through more than one refold.

Pre-oracle transforms

Reading outputs Finding a region in a generated structure by its standard annotations, without hand-parsing CIFs. FAQ Common questions and gotchas across condition specs, generation, and metrics.

More guides are coming; this page will grow with them.