Optimiser Repairs for Living-Cell Self-Driving Laboratories: Performance and Redundancy under Experimental Constraints

Living-cell self-driving laboratories face challenges in optimizing experiments due to delayed readouts, changing responses, and limited sample availability. In simulations, researchers assessed when optimiser repairs enhance closed-loop performance and when platform constraints negate their need, distinguishing sample consumption from permanent loss of repeatable conditions.

The study confirms that exclusions and confirmation requests yield the baseline trajectory only when acting on conditions already prohibited by the platform, sharing the same posterior and selection procedure, and releasing infeasible requested slots. This conclusion applies to pending-exclusion, previous-batch, and confirmation rules under permanent condition retirement but does not address arbitrary selection interventions.

To validate their findings, the team conducted 33,900 simulated policy runs with 200-300 paired campaign seeds per setting. Across 27 landscape, assay-size, and kernel settings, previous-batch exclusion modified final normalized performance by -0.004 to +0.283, with positive differences remaining statistically significant in 13 settings after Holm adjustment.

These effects are minimal on rugged landscapes. Permanent retirement can be detrimental when dealing with noisy, rugged responses. The results emphasize the significance of explicit availability constraints and evaluation across various experimental conditions when examining optimiser repairs.