back to list
Project: Cheaper drugs by auto-optimizing photo-chemistry
Description
Photo-chemistry is a technique where input chemicals are first ionised and then new molecules are synthesised through interactions with photons.
The exact amount of input chemicals is very important to make such reactions run effectively. Currently, Bayesian Optimization is used to find the optimal mixtures, but since every reaction takes time, it can also take some time before the best mixture is found.
To make this more efficient, there are several ideas we can try:
* Defining a prior inspired by chemistry knowledge: https://arxiv.org/abs/2006.14608
This can be a simple Gaussian distribution around the most likely value for their input parameters. i.e. the most likely 'limiting reagent' or 'residence time'.
* Transferring surrogate models from previous experiments: https://arxiv.org/abs/1802.02219 This is useful since we already have the surrogate models from many prior experiments. We only need to store them and then reuse them to optimize their process parameters in much fewer steps.
Other ideas are certainly valuable. The goal is to ultimately implement and evaluate (and possibly combine and adapt) these ideas in a real-world photochemistry installation at the university of Amsterdam.
Details
- Supervisor
-
Joaquin Vanschoren
- Interested?
- Get in contact