Abstract:
Pharmaceutical research and development harnesses chemical tools to positively impact human health, but it is also an extremely costly and risky process. With increasing amounts of available data, machine learning and data science are poised to accelerate and de-risk this process by effectively prioritizing molecular hypothesis and streamlining biochemical testing. In this talk, I will introduce multiple applications of machine learning to support the different stages of the drug discovery and development process. Specifically, I will showcase opportunities for using machine learning in de novo drug design and for predicting biological targets of small molecules. [1-3] I will briefly discuss how such predictive models can be integrated into automated workflows to guide medicinal chemistry [4, 5] and how they can be adapted for reaction condition optimization. [6] In the second half of the talk, I will present a novel pipeline that enables the delivery of poorly soluble drugs through self-assembled nanoparticles. [7] Assisted by laboratory automation and machine learning, I developed and in vivo validated sorafenib nanoparticles to treat hepatocellular carcinoma. [7] Finally, our pharmaceutical data analysis showcases that inactive ingredients might have a larger impact on health outcomes than previously anticipated [8] and my current work is focused on developing innovative machine learning tools to guide such discoveries for personalized medicine. [9]
[1] Reker et al. Proc. Natl. Acad. Sci. USA 111, 4067–72 (2014).
[2] Reker et al. Nat. Chem. 6, 1072–8 (2014).
[3] Reker et al. Angew. Chem. Int. Ed. 53, 7079–84 (2014).
[4] Reker et al. Drug Discov. Today 20, 458–465 (2015).
[5] Reker et al. Chem. Sci. 7, 3919–27 (2016).
[6] Reker et al. ChemRxiv (2018).
[7] Reker et al. BioRxiv (2019). https://doi.org/10.1101/786251
[8] Reker et al. Sci. Transl. Med. 11, eaau6753 (2019).
[9] Reker et al. in preparation.