Boltz compatibility example

Boltz Lab 1.1.0 shows how to profile atomic scientific values while leaving operational and aggregate records unstandardized.

Port	Direction	Profile	Why
protein_sequence	Input	protein.sequence/v1	One amino-acid sequence
ligand_smiles	Input	chemical.smiles/v1	One molecular SMILES string
msa_path	Input	protein.multiple-sequence-alignment/v1	One A3M file
run_options	Input	Unstandardized	Runtime configuration
binding_probability	Output	boltz.binding-probability/v1	Atomic Boltz binder probability
affinity_log10_ic50_micromolar	Output	boltz.log10-ic50-micromolar/v1	Atomic transformed affinity
predicted_structure	Output	protein-ligand.complex-structure-mmcif/v1	One mmCIF structure artifact
affinity_summary	Output	Unstandardized	Multi-field Boltz aggregate retained for consumers
confidence_summary	Output	Unstandardized	Boltz-specific confidence aggregate
structure_artifacts	Output	Unstandardized	Operational artifact collection
run_metadata	Output	Unstandardized	Command, logs, status, cache, and provenance

Boltz reports affinity_probability_binary as a binder probability in the inclusive range 0 to 1. The wrapper exposes it as binding_probability.

Boltz reports affinity_pred_value as log10(IC50) where IC50 is expressed in micromolar. Lower values imply stronger predicted affinity. It is not pIC50 and must not be compared with a different concentration basis or logarithmic convention without an explicit adapter.

The standardized structure output is mmCIF only. The A3M profile verifies the file’s basic representation; the wrapper separately verifies that its query sequence matches protein_sequence. None of these checks establishes binding, affinity, alignment quality, or structural accuracy as experimental fact.