Guiding Drug Optimisation Using Deep Learning Imputation and Compound Generation
International Pharmaceutical Industry : Summer 2020 Vol 12:2
The use of machine learning (ML) methods is now commonplace in many disciplines and artificial intelligence (AI) is on the rise, promising better and smarter solutions to ‘all your problems’. However, despite the hype, there is increasing evidence we have entered the next ‘AI winter’ or the so-called ‘trough of disillusionment’ in the ongoing hype cycle1. There is still a gap in understanding on the route from traditional and well-understood statistical modelling methods to the poorly-defined promises of AI, and exactly how the majority of researchers can cross that gap is not clear.
Researchers in drug discovery are familiar with quantitative structure activity relationship (QSAR) model building methods. Many of these methods now employ forms of machine learning (ML), a sophisticated form of ‘fitting functions to data’. The question is how to leap forward from this well-known and comfortable ML space toward sophisticated AI tools, by which we mean: A connected set of ML components in an automated system which together produce a rich behaviour capable of solving complex tasks. The lesser known ‘AI’ is augmented intelligence, and there is no reason why a human cannot be part of the connected components in this sophisticated AI system. The combination of a human expert and superior tools has been found to be optimal as well as convenient.
You can download this article as a PDF