The ‘design, make, test’ cycle is central to medicinal chemistry.  While effective, the iterative nature of this cycle can drive inefficiency and protracted optimization timelines. Coupled with evolving multi-parameter optimization needs, the medicinal chemist is challenged to assimilate a large quantity of diverse data and to find solutions to diverse problems. Instead, a ‘design, <i>predict</i>, make, test’ cycle could drive greater efficiency and shorten timelines by focusing on collecting the necessary data to make the next set of informed predictions. This requires a new data-driven decision-making culture, ensuring that tools, processes, and people all have modelling and analytics available as a key component from the beginning. This talk will explore how machine learning has been applied to all stages of the design-predict-make-test cycle, leveraging historical data to generate synthetically tractable molecules at pace that mitigate ADMET and scalability risks.  We will also discuss design platforms that encourage the incorporation of predictive methods and democratization of these models to all chemists, with the limitations and lessons learned from deploying these approaches on real projects.

Designing for developability: Machine learning in data-driven drug discovery

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.