Web-BO

A GUI to support data-driven optimisation for chemical tasks

key features • examples • run locally • citation • acknowledgements • license

video demonstration

key features

Web-BO is a web application offering a graphical user interface (GUI) to facilitate the application of BO to a wide array of chemical problems.

Users define the optimisation task and perform the measurements, and Web-BO takes care of the rest!

Web-BO offers several key features:

modular platform that is easily integrated into existing ELN frameworks
can be used as a standalone database and optimiser for chemical tasks. All of the data is stored in a searchable SQL database in an intuitive and future-proof form, enabling efficient benchmarking of varying data-driven optimisation algorithms.
no coding experience is necessary to interface with and apply BO algorithms to chemical optimisation applications.
can integrate bespoke BO solvers
optimisation space validation to ensure seamless integration

examples

Web-BO possesses 2 examples and 1 walk-through; one example demonstrates a sample optimisation task (described in the manuscript and below), one example provides sample input csv files, and the walk-through demonstrates the integration with datalab.

optimisation task demonstration

To help you get familiar with all that Web-BO has to offer, we provide an optimization case study that takes advantage of existing reaction emulators to acquire the experimental measurements.

This case study concerns optimizing the coupling of 3-bromoquinoline with 3,5-dimethylisoxazole-4-boronic acid pinacol ester in the presence of 1,8-diazobicyclo[5.4.0]undec-7-ene (DBU) and THF/water, Figure 1.

Figure 1. Suzuki-Miyaura cross-coupling reaction involved in this case study.

The parameter space involved in this optimisation task is detailed in Figure 2.

Figure 2. Parameter space spanned by this case study includes 3 continuous variables (catalyst loading, temperature, and residence time) and 1 categorical variable (catalyst).

The catalyst options are presented in Figure 3.

Figure 3. The catalyst options included in this study.

This case study takes advantage of the experiment emulators offered by the Summit package. Specifically, the Suzuki-Miyaura Cross Coupling Emulator.

toy dataset demonstration

We also offer an example of the csv format expected by Web-BO; this is in the toy-datset directory. This is comprised of a dataset and a list of categories for the categorical variable.

solvents	temperature	pressure	target
a	25	1.0	2.344
b	45	2.2	4.5555
c	25	2.3	10.333
d	55	2.4	1.3
e	10	2.0	4.5
f	75	1.0	2.455
g	65	1.5	3.0
categorical	integer	continuous	continuous

The solvents are a categorical variable. Thus, we need to define all of the options that comprise the optimisation space. This takes the general form:

solvents
a
b
c
...
x
y
z

In order to use the toy dataset, these must first be downloaded locally and then uploaded using the csv upload feature in Web-BO.

run locally

To deploy Web-BO locally, follow the below steps:

foo@bar:~$ git clone https://github.com/austin-mroz/webBO
foo@bar:~$ cd webBO
foo@bar:~/webBO$ conda env create -p ./.venv -f environment.yml
foo@bar:~/webBO$ conda activate ./.venv
foo@bar:~/webBO$ python main.py

Web-BO will be hosted on your localhost using its default port (8050); http://localhost:8050/

citation

A. M. Mroz, P. N. Toka, E. A. del Rio Chanona, K. E. Jelfs, submitted.

acknowledgements

The authors thank Dr. Matthew Evans, one of the developers of datalab for insightful discussions and assistance in working with datalab's API. The authors would also like to thank Dr. Diego Alonso Alvarez and Dr. Lukas Turcani for their technical assistance in making Web-BO publically available. A.M.M. is supported by the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, a Schmidt Futures program. K.E.J acknowledges the European Research Council through Agreement No. 758370 (ERC-StG-PE5-CoMMaD) and the Royal Society for a University Research Fellowship.

license

We release this software under the conditions of the MIT license.

ImperialCollegeLondon/webBO