SLAS 2019 AI in Process Automation Symposium

Applications of Artificial Intelligence in Drug Discovery – Separating Hype From Utility
Artificial intelligence and machine learning are now impacting many aspects of the drug discovery process. One factor driving the rapid adoption of machine learning in drug discovery is the availability of software tools for building machine learning models. Software that was once only available to machine learning researchers is now freely available for download. Organizations such as Facebook, Google, and Amazon are now releasing Open Source software tools for machine learning that can be applied to problems in numerous domains including drug discovery. While these new tools offer an array of opportunities, we must be adequately prepared to use them effectively. Researchers need to have a sufficient command of the underlying science to develop representations that can be processed by machine learning algorithms. Scientists applying machine learning in drug discovery must have sufficient computational skills to be able to adopt and apply these techniques. Finally, researchers must have an understanding of statistics that will enable them to assess and improve the quality of their models.

Ultimately. the success of any predictive model comes down to three factors; data, representation, and algorithms. The data being generated must be relevant, and the experimental error must be well characterized. In order to build a predictive model, we must be able to create a representation of the data, typically as some sort of vector, that can be processed by a machine learning algorithm. The algorithm can then identify relationships between the data and some observable (e.g. biological activity) and can subsequently be used to make predictions on new data.