Abstract:
In recent days, machine learning has been increasingly adopted for accelerating in silicon drug discovery. However, because of difficulties in obtaining qualified drug-related data, prediction models are vulnerable to over-fitting problems and lack generalization ability. Also, some ML models may not be useful in real-world applications in that models lack domain knowledge such as well-defined chemical synthesis route and underlying physics in molecular interaction. To bridge the gap between ML researchers and domain experts, computational chemistry approaches have been utilized in the aforementioned applications. In this talk, I will review current research trends in drug discovery with machine learning and computational chemistry.
Firstly, I will briefly introduce my past ML research in applications of graph neural networks and Bayesian learning for molecular property predictions.Then, I will present high-throughput virtual screening of commercially available libraries designed with named chemical reactions.
Since screening the entire library is extremely cost-expensive, an active learning approach is combined to reduce computational efforts. Lastly, I will introduce retrosynthetic reaction analysis with ML models, which is widely used in the lead optimization step.

Bio:
Dr. Ryu is a research scientist at GALUX, a company aiming to accelerate drug discovery with AI and computational chemistry. Prior to that, he worked at AITRICS as a research scientist after completing his Ph.D course in 2020 from the department of chemistry at KAIST. His research mainly focuses on small molecule drug discovery with computational chemistry tools, including molecular docking, molecular dynamics and in silicon reaction planning, and machine learning, especially graph neural networks, Bayesian learning and active learning.

Zoom link : https://zoom.us/j/6906724188?pwd=QVI3Tkp0M3FnWHhodXhKY2NZSUx5Zz09

POSTECH ML Winter Seminar Series Homepage : https://sites.google.com/view/pair-ml-winter-seminar-2022/home