Big data and deep learning: extracting and revising chemical knowledge from data

What is it about?

The recent advancements of machine learning are making it possible to create systems that automatically mine patterns and learn from data. In chemistry, in particular drug discovery and properties prediction, machine learning is changing the way of creating and testing new molecules. After presenting the principal learning methods, focusing on neural networks and deep learning, a study case is presented. Using a large data set of molecules data tested for the mutagenicity property, three models for predicting this property are derived from data, considering the molecules as graphs, as text, or as images. The knowledge extracted from the networks is analyzed and positively compared with the toxicity rules for mutagenicity.

Featured Image

Photo by Vedrana Filipović on Unsplash

Photo by Vedrana Filipović on Unsplash

Why is it important?

There major challenges are considered. • One challenge is to improve the process of constructing predictive models for chemistry. • Another challenge is to avoid, or to reduce, the bias of the expert in modeling. • The last challenge is to provide an explanation of the result obtained by the model.

Perspectives