Scientists working with formulated products share some common challenges that supersede the application or industry designation. In particular, the challenge of deformulation pervades the consumer goods industry, wherein a product’s ingredients are largely known because of regulations requiring labeling, however their quantities in the formulated product are not known. In this work, CAS and Citrine Informatics aim to achieve predictive deformulation using a variational autoencoder method, applied to a subset of CAS’s collection of over 5 million scientist-curated formulated product recipes. With this method, deformulation is achieved with significantly more accuracy than a simpler nearest-neighbor approach.

Predictive chemical deformulation: A deep generative approach

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.