I have been given some data on a biotechnological process stemming from a multitude of different (kind of independent) experiments. The process has a lot of potentially relevant inputs (temperature, medium composition, microcarrier concentration, cell type and many more) and one of the goals is to maximize product yield by finding out which factors/interactions are relevant and what their optimal value is. Currently it seems to me like a Response Surface Modelling approach would be most appropriate for that. Another goal is to find out where there are "gaps" in the space covered by the experiments so far conducted, i.e. which additional experiments would best help discover the role of each factor/interaction or most reduce uncertainty in the model.

As you can see, I don't even know how to properly express this in statistical terms which makes it hard to google for solutions, but does anyone understand what I'm aiming for and is able to point me in the right direction? Bonus points if you know of an R or python package that can help me actually compute it.


You'll probably find what you're looking for if you read up on bayesian optimisation for experimental design. It's a formalism that describes how you can maximally reduce uncertainty. It tells you which regions in parameter space will, if measured, restrict the uncertainty in your model the most.

I found a recent review article here: https://ieeexplore.ieee.org/abstract/document/8957442

| cite | improve this answer | |
  • $\begingroup$ That's incredibly interesting, thank you so much! And there's even libraries linked at the end of the paper. <3 It looks like that is exactly what I was looking for. $\endgroup$ – DP. Jun 4 at 12:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.