What is it about?
Deep neural networks are effective in learning directly from raw data without the need of feature extraction. This paper shows how QSAR models can be constructed from SMILES without computing chemical descriptors. The case presented is about predicting the output of the Ames test for mutagenicity. The results positively compare with the state of the art models.
Photo by Myriam Jessier on Unsplash
Why is it important?
For the first time we show a process for building performant QSAR models using deep neural models based on recurrent networks. The values in the last layers of the trained network represent the important substructures found by the net and constitute an explanation of the QSAR result.
Read the Original
This page is a summary of: Could deep learning in neural networks improve the QSAR models?, SAR and QSAR in Environmental Research, August 2019, Taylor & Francis,
You can read the full text:
The following have contributed to this page