Open Source Bayesian Models. 1. Application to ADME/Tox and Drug Discovery Datasets

  • Alex M. Clark, Krishna Dole, Anna Coulon-Spektor, Andrew McNutt, George Grass, Joel S. Freundlich, Robert C. Reynolds, Sean Ekins
  • Journal of Chemical Information and Computer Sciences, June 2015, American Chemical Society (ACS)
  • DOI: 10.1021/acs.jcim.5b00143

Using open source Bayesian models for drug discovery

What is it about?

This work describes how open source FCFP6 descriptors and Bayesian algorithm are implemented in the CDD Vault. Various ADME/Tox and drug discovery models are built to demonstrate the utility. In addition models are built in other software and implemented in mobile apps such as MMDS.

Why is it important?

We demonstrate an approach using open source tools inside a commercial tool that can be used to export models and share them or enable secure sharing in the CDD Vault, This work is important as most of the technologies were not open source. The ability to share models represents a step forward for making models more accessible.


Dr Sean Ekins
Collaborations in Chemistry

This work builds on earlier papers in 2010 that showed open source descriptors could build equivalent models to commercial tools. It also extends the use of the fingerprint descriptors described previously in the context of TB Mobile Version 2.0. With the companion paper we describe how the FCFP6 and ECFP6 descriptors and Bayesian algorithm could be used to build thousands of models relevant to drug discovery with good ROC values

Alex Michael Clark
Molecular Materials Informatics

The paper builds on previous open source work adding ECFP6 and FCFP6 structure-based fingerprints to the CDK toolkit. It describes in great detail how the Bayesian model building algorithm is implemented, with the intention of ensuring that anyone can reimplement it. A number of studies are included to show that the method is just as effective as commercial products. The overall purpose of the article and the public release of the software is to encourage model sharing, which is currently relatively rare, much to the detriment of the drug discovery community.

Read Publication

The following have contributed to this page: Dr Sean Ekins and Alex Michael Clark