Open Source Bayesian Models. 3. Composite Models for Prediction of Binned Responses

  • Alex M. Clark, Krishna Dole, Sean Ekins
  • Journal of Chemical Information and Computer Sciences, February 2016, American Chemical Society (ACS)
  • DOI: 10.1021/acs.jcim.5b00555

Semi quantitative open source Bayesian models

What is it about?

This work builds on the previous 2 papers on open source Bayesian Models by us to describe a new method for creating composite models - data has multiple bins and we attempt to predict these instead of binary bins.

Why is it important?

We wanted to expand the scope of Bayesian models, make them more accessible to non-experts. But at the same time binary data may not be ideal when the underlying data is high quality dose response data, so an expansion of this approach was warranted. Multiple bins provides a semi-quantitative approach to prediction. Multiple Bayesian models essentially compete and the predicted bin is from the model with the highest score.


Dr Sean Ekins
Collaborations in Chemistry

This paper makes use of the ChEMBL and other datasets which we have recently published on. All the underlying algorithms are on Github. The work builds on a long standing interest in machine learning and how we can make these approaches more accessible and user friendly.

Read Publication

The following have contributed to this page: Dr Sean Ekins