This page summarizes our current method development directions.
AI-driven Ligand docking at an incomplete context
Drug discovery scenario frequently happens at an incomplete receptor structure context. We study how to select & dock drug compounds into those receptor structures (e.g. AlphaFold models) using Deep Learning techniques.
Related works
H. Park et al, PNAS 2018.
H. Park et al, JCTC 2020.
N. Hiranuma & H. Park et al, Nat Commun. 2020
Funding
NRF Korea, Starting Grant 우수신진연구 (2022~2024)
Unsupervised Learning of Receptor-Drug matching
Data sparsity is always a bottleneck for applying DL on protein-ligand interactions. Our goal is to learn a general chemistry-aware AI called MotifNet on Protein-protein interface data instead, and to transfer the knowledge to protein-ligand interface systems for virtual screening (VS) and target receptor identification applications.
Related works
MotifNet, under preparation.
H. Park et al, JCTC 2020.
J. Dou & A. Vorobieva et al, Nature 2018.
De novo Peptide Drug discovery
Peptides are attractive drug molecules. We aim to develop a general Deep Learning framework for peptide drug discovery based on chemical principles (i.e. receptor structure alone). Peptide backbone & sequences are designed to optimally satisfying the predictions from aforementioned MotifNet AI.
This work is a collaboration with SNU Seok lab.
Related works
MotifNet, under preparation.