torchcell.metabolism.parameters.KcatPredictor

class KcatPredictor(*args, **kwargs)

Bases: Protocol

A sequence-based turnover predictor.

Deliberately minimal. Any of KcatNet, RealKcat or DEKP satisfies it, and a predictor that also emits \(K_M\) advertises that through emits_km so the caller can close the affinity gap in the same pass rather than running two models.

predict(sequence: str, substrate_smiles: str | None = None) → tuple[float, float | None]

Return (kcat_per_s, km_mM_or_None) for one (protein, substrate) pair.