Smart design of organic and organometallic materials using automated machine learning methods and workflows

H.S. Kwak, T. Robertson, C.M. Krauter, K. Leswing, M.D. Halls
Schrodinger, Inc.,
United States

Keywords: QSPR, automation, deep learning, de novo design


Data-driven design and optimization of highly efficient organic and organometallic materials has recently gained extensive interest in a wide variety of materials and chemical industry applications. In this work, we showcase the latest technology in machine learning and informatics-driven ideation and design solution developed by Schrödinger. The technology is highlighted by cutting-edge, deep-neural-network solutions for the de novo design of small molecules with desired properties, as well as automated machine-learning workflow for predicting physico-chemical properties of molecular compounds in high-throughput fashion via qualitative structure-property relationships (QSPR). Examples of materials screening for various sets of applications, including OLED materials design, ALD precursor screening, and clean fuel research will be demonstrated. The automated data-driven predictive scheme provides unbiased measures to quickly assess the key design rules in chemical space, which could significantly lower the barrier towards developing novel materials solutions.