Back

Conversational Chemistry: A Novel Approach to Chemical Search and Property Prediction

Ben-Joseph, J.; Oates, T.

2023-11-15 bioinformatics
10.1101/2023.11.11.566721 bioRxiv
Show abstract

We have developed an approach to train a chemical property prediction model using both English and the SELFIES chemical language describing the structure of small, drug-like molecules. This model generates chemical embedding vectors, which we then use to train classification models. Our straightforward softmax classification model surpasses the commonly-used message passing neural network architecture in certain chemical property prediction tasks. Moreover, these chemical embedding vectors can be employed in other applications, such as building a chemical search engine that enables users to find new drugs with natural language queries (e.g., "low toxicity blood brain barrier permeable drug that inhibits HIV replication").

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.