Machine Intelligence for Efficient Large-Scale Reaction Optimisation with Automation
minerva is a repository containing code for the application of scalable, multi-objective, batched reaction optimisation described in the manuscript associated with this work. From a user-defined space of reaction conditions, minerva initialises the optimisation campaign with quasi-random Sobol sampling, and performs large-scale batch optimisation of chemical reactions. minerva has been deployed experimentally with search spaces of ~100,000 with batch sizes of 96 in High-Throughput Experimentation.
git clone https://github.com/schwallergroup/minerva.git
cd minerva
conda create --name minerva python=3.8
conda activate minerva
pip install -r requirements.txt
pip install -e .
Installation will require several minutes. This method was tested and developed on CUDA-enabled GPUs (linux OS) with CUDA = 11.7 using the install below.
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia
In the notebooks directory, there are 5 Jupyter notebook tutorials. These tutorials detail how to run benchmarks on virtual datasets, create a reaction condition space, initialise and run iterations of optimisation with collected experimental data.
In the scripts directory, we include execution scripts used to run optimisation rounds and obtain optimiser suggestions on the experimental ML HTE campaign described in the manuscript. All computations in the manuscript were run on a workstation with an AMD Ryzen 9 5900X 12-Core CPU and a RTX 3090 (24GB) GPU.
All HTE experimental data collected from experiments described in the manuscript are available in the experimental_campaigns/experiments/publication/SURF_files directory in the Simple User-Friendly Reaction Format (SURF).