The abstract is read at the publisher; the summary is JClub's.
Surrogate verifiers trained on N=7,688 Lacticaseibacillus growth curves disagree on which media experiments to test next, and model choice dominates Bayesian optimization outcomes.
- Why it matters: Closed-loop labs rely on surrogate models instead of experimental ground truth, but it is unclear how surrogate accuracy, uncertainty, and extrapolation affect which candidates get expensive follow-up.
- What they did: The study trains six surrogate families (polynomial, Gaussian-process, tree-ensemble, neural, and Monod-kinetic) to predict maximum OD600, then tests them in a Bayesian optimization loop while applying conformal calibration for uncertainty and coverage.
- The result: In transfer to an unseen related bacterial strain and on a prospective validation plate, conformal intervals fail in the direction and region predicted by retrospective analysis, implying calibration can break under strain shift.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.