The code for the BERT Enriched Embedding model, for yield prediction.

git clone https://github.com/PauloNeves-Git/BEE.git
cd BEE
conda env create -f bee38.yml
pip install --no-cache torch==1.8.2+cu111 torchvision==0.9.2+cu111 torchaudio==0.8.2 -f https://download.pytorch.org/whl/lts/1.8/torch_lts.html

This code relates to a pre-print available on Chemarxiv https://chemrxiv.org/engage/chemrxiv/article-details/633c44a0ea6a224e2408b7cd.
The paper discusses the dramatic impact that yield prediction models can already have in the Pharmaceutical Industry and how to improve them, for this reason most results presented in the publication requiere access to industry-scale datasets which are private. For the sake of reproducibility a notebook called "BERT Enriched Embedding w Open-Source Data" was created to reproduce the results with open source data.

More notebooks which will enable open-source reproduction of others experiments (uncertainty estimation plots) on the paper will be added.

• https://github.com/rxn4chemistry/rxnfp
• https://github.com/ThilinaRajapakse/simpletransformers
• https://github.com/huggingface