Cardiotoxicity has been a leading concern behind the recall of several marketed drugs and failure of clinical candidates. Inhibition of the potassium ion channel whose alpha subunit is encoded by human Ether-à-go-go-Related Gene (hERG) leads to prolonged QT interval in the cardiac action potential. Fatal cases of cardiac arrhythmia have been reported due to drug-induced hERG channel inhibition. Several computational techniques, including machine learning (ML), have been employed to develop QSAR models for hERG. However, models based on artificial intelligence (AI) methods have been rarely reported. To this end, we performed a comprehensive comparison of binary classification models developed using the state-of-the-art ML and deep learning methods employing molecular fingerprints, whole molecular descriptors, SMILES and latent descriptors. Latent descriptors are numerical descriptors derived from the latent space of the autoencoder and adversarial autoencoder models built using all small molecules available at ChEMBL and PubChem databases. The training data (~6000 compounds) for the QSAR model was extracted from ChEMBL while the validation data (~1800 compounds) was generated internally using FluxORTM K+ channel assay. The latter set was pooled from an in-house compound library which was included while building the encoder models. We noticed that the performance of deep learning models was on par with the state-of-the-art ML method (Random Forests). Interestingly, the models based on latent descriptors performed well on validation data. Further, we compared our models with the models based on graph convolutional networks (GCN) that use molecular graphs as features. All our models outperformed the GCN models on validation data. Overall, this study highlights the application of AI methods to chemistry big data, readily available from publicly-accessible compound databases, for generation of novel descriptors that are promising for QSAR modeling. However, at least in case of cardiotoxicity, we demonstrate that such models do not add a significant advantage over ML models based on conventional descriptors. We believe a large benchmark study on diverse drug discovery data sets will validate our findings.

Machine learning vs. AI methods for modeling cardiotoxicity in the ‘big data’ era

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

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