What is it about?

The paper reviews the concept of similarity in philosophy and mathematics, and examines its role in QSAR models. After discussing the similarity paradox in QSAR, it is shown how similarity is implicitly used by deep learning models when building and using the auto-generated features. Data about mutagenicity prediction are used in the discussion.

Featured Image

Why is it important?

It is shown that the QSAR models generated by deep learning from the whole molecules structures are even more accurate than models generated from pre-computed molecular descriptors; moreover, they are more robust against the similarity paradox.


I hope this article can open a bridge between classical chemometrics and machine learning.

Prof Giuseppina Carla Gini
Politecnico di Milano

Read the Original

This page is a summary of: The QSAR similarity principle in the deep learning era: Confirmation or revision?, Foundations of Chemistry, July 2020, Springer Science + Business Media,
DOI: 10.1007/s10698-020-09380-6.
You can read the full text:




The following have contributed to this page