Name of the participant: Maximilian Harl
Description of the IT research project: Drug design, especially drug development, is generally very costly in terms of resources and time, with costs of up to 2.6 billion euros and time horizons of up to 20 years. For this reason, several studies have investigated the usefulness of various machine learning methods for drug design and have shown that, compared to classical methods, they can predict many properties of molecules both quickly and relatively accurately [1, 2].
This research project deals with the investigation of reinforcement learning methods in drug design and process management (esp. process mining). In these methods, possible decisions are analyzed with estimates of their effects in order to approximate a behavior that pursues the desired goal. In the case of drug design, these are desired physiochemical or biological properties. With the contemporary advances in Deep Learning, there is also a renaissance of Reinforcement Learning, especially in combination of both techniques. In the context of this project, reinforcement learning offers several advantages, first of all, that the modifications of molecular graphs can be used as a basis for reinforcement learning methods, which also makes only chemically valid modifications [3].
This project will investigate how molecules can be optimally represented for reinforcement learning methods and how reward functions for drug design can be formulated in a target-oriented manner. Furthermore, these (molecule) graphs can be used in the context of process management (esp. process mining) for modeling business processes. For this reason, the potential of drug design methods in the domain of process management (esp. process mining) and how they can be applied in both domains will also be investigated.
Software Campus partners: FAU, Merck
Implementation period: 01.03.2021 – 28.02.2022
[1] Elton, D. C., Boukouvalas, Z., Fuge, M. D., & Chung, P. W. (2019). Deep learning for molecular design—a review of the state of the art. Molecular Systems Design & Engineering, 4(4), 828-849.
[2] Zhavoronkov, A., Ivanenkov, Y. A., Aliper, A., Veselov, M. S., Aladinskiy, V. A., Aladinskaya, A. V., … & Aspuru-Guzik, A. (2019). Deep learning enables rapid identification of potent DDR1 kinase inhibitors. Nature biotechnology, 37(9), 1038-1040.
[3] Zhou, Z., Kearnes, S., Li, L., Zare, R. N., & Riley, P. (2019). Optimization of molecules via deep reinforcement learning. Scientific reports, 9(1), 1-10.
