webinar register page

Combining High-throughput Atomic Scale Simulation and Deep Reinforcement Learning in the Discovery of Novel OLED Materials with Targeted Optoelectronic Properties
Hole-transporting materials (HTM) are a critical class of organic semiconductors, required for the fabrication of a variety of state-of-the-art display and semiconductor devices. In this work, we apply the technique of using Recurrent Neural Networks (RNN) as a generative model on SMILES representation of molecules, which has demonstrated success in drug discovery, to design HTMs targeting specific properties. A set of training compounds were selected from OLED material chemical space expanded from commercial catalogs to run optoelectronic properties calculations using Quantum Mechanics (QM) tools from Schrӧdinger’s Materials Science Suite. Reinforcement learning was used to fine-tune pre-trained RNN based on both the prior likelihood and a scoring function defined by the user, thus allowing optimization of multiple properties at once, which has been a huge challenge in material design. We demonstrated that by carefully defining the applicable chemical space, and providing accurate physics-based property calculation on a large number of compounds, our data-driven approach, with the aid of advanced machine learning techniques, can be expanded to many different domains to systematically discover novel materials with targeted properties.

高通量原子尺度计算模拟与深度强化学习结合应用于具有针对光电特性的新型OLED材料的研发
空穴传输材料(HTM)作为有机半导体的关键类别是制造各种先进的显示器和半导体器件所必需的。在这项研究中,我们应用递归神经网络(RNN)作用于分子的SMILES的技术(该技术已在药物研发中有所成绩),来设计具有针对特定光电特性的HTM。我们从OLED材料化学领域(商业目录扩展而来)中选择了一组训练样例化合物,使用Schrӧdinger材料科学软件套件中的量子力学(QM)工具进行光电性能计算。强化学习根据先验可能性和用户定义的函数对预先训练的RNN进行微调,从而允许一次优化多个属性,这在材料设计中一直是一项巨大的挑战。我们的研究证明,通过选用合适的化学空间,并提供大量准确的基于物理学的化合物的性质计算数据,借助于先进的机器学习技术,我们的数据驱动的方法可以扩展到许多不同的领域,从而系统地发现具有针对性能的新型材料。

Apr 7, 2021 09:00 AM in Beijing, Shanghai

Webinar logo
Webinar is over, you cannot register now. If you have any questions, please contact Webinar host: Marketing.