Machine Learning and Artificial Intelligence techniques are increasingly used in ongoing drug discovery projects at Merck KGaA, Darmstadt, Germany. We present how Machine Learning and Deep Learning models are utilized for target identification and optimization of protein kinase inhibitors with respect to potency, selectivity and ADME profile (e.g. microsomal stability, permeability, and brain penetration). For prediction of kinase selectivity, we employ more than 270 deep learning models, which help medicinal chemists in the selection of the right molecules to synthesize. We will also share lessons learned from analysis of large protein kinase profiling datasets. Finally, three challenges for predictive modelling in drug discovery projects will be addressed: a) how the quality and applicability of predictive models can be assessed, b) how models are interpreted, and chemists can be guided by mapping information from Deep Networks back onto molecular structures and c) how to best operationally integrate predictive models into project work to maximize impact.

What compound to synthesize next? How machine learning and artificial intelligence impact compound optimization

2019 Fall National Meeting

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