Deep learning is currently the most popular technique of machine learning, and every other day you read new success stories about how it is dramatically changing a wide range of applications. In recent years, it has been increasingly used in drug discovery to predict potential drug targets, screen for active molecules or plan chemical retrosynthesis routes. In this talk I will give a brief insight into how we use deep learning in our research to (i) de novo design molecules under multi-parametric objectives, (ii) predict protein-ligand interactions using unsupervised learned representations, (iii) solve the inverse molecular problem for extended chemical fingerprints or to (iv) predict molecular properties with multitask graph convolutional networks.
Director Machine Learning Research, Bayer AG
Djork-Arné Clevert joined the Bayer AG in 2015 and is since May 2019 the Director of the Machine Learning Research department at Bayer AG. He has a background in computer science and received his doctorate on machine learning for computational biology. He was a senior scientist in the prestigious Hochreiter Lab at the Institute of Bioinformatics at Johannes Kepler University from 2007 to 2015. He has been Co-PI in several large projects with pharma industry, in particular, transcriptome analysis and statistical genetics with Johnson and Johnson and Merck Serono Geneva, respectively. His research has mainly been concerned with microarray data in earlier years. Later he shifted this research focus to the prediction of biological effects of compounds with methods, such as, deep neural networks. A highlight of his Marie Curie fellowship was the introduction of the Exponential Linear Units (ELUs), which has become a defacto standard in the field of deep learning. He is promoter of multiple master and doctoral projects as well as of 6 completed postdoctoral projects. He is author of more than 40
publications in international peer-reviewed journals and books.