The idea
When each experiment consumes material, time or equipment capacity, choosing the next test becomes a decision problem. BoFire combines Bayesian optimisation with other experimental-design strategies in an open-source Python framework aimed at practical chemistry.
What the authors established
The paper emphasises configurable problem descriptions and the engineering work needed to move research methods into industrial workflows. Its serialisable formulations support integration with automated laboratories and human-in-the-loop systems.
Applications
Formulation development and process tuning benefit from selecting experiments under a fixed budget. Sequential policies can balance improvements in the objective with the information gained from each test.
Where care is needed
A framework does not establish a universal reduction in the number of experiments. Measurement quality, domain constraints and safe operating ranges need expert review.
Sources and credits
- BoFire: Bayesian Optimization Framework Intended for Real Experiments
Johannes P. Dürholt, Thomas S. Asche, Johanna Kleinekorte, Gabriel Mancino-Ball, Benjamin Schiller, Simon Sung, Julian Keupp, Aaron Osburg, Toby Boyne, Ruth Misener, Rosona Eldred, Chrysoula Kappatou, Robert M. Lee, Dominik Linzner, Wagner Steuer Costa, David Walz, Niklas Wulkow, Behrang Shafei, published
Journal of Machine Learning Research 26(204), 1–7 (2025).
BoFire is the cited authors’ framework. The separately credited industrial photograph provides context.