Powerful, transferable representations for molecules through intelligent task selection in deep multitask networks

  • 2018-09-17 17:06:06
  • Clyde Fare, Lukas Turcani, Edward O. Pyzer-Knapp
  • 6

Abstract

Chemical representations derived from deep learning are emerging as apowerful tool in areas such as drug discovery and materials innovation.Currently, this methodology has three major limitations - the cost ofrepresentation generation, risk of inherited bias, and the requirement forlarge amounts of data. We propose the use of multi-task learning in tandem withtransfer learning to address these limitations directly. In order to avoidintroducing unknown bias into multi-task learning through the task selectionitself, we calculate task similarity through pairwise task affinity, and usethis measure to programmatically select tasks. We test this methodology onseveral real-world data sets to demonstrate its potential for execution incomplex and low-data environments. Finally, we utilise the task similarity tofurther probe the expressiveness of the learned representation through acomparison to a commonly used cheminformatics fingerprint, and show that thedeep representation is able to capture more expressive task-based information.

 

Quick Read (beta)

loading the full paper ...