Research Interests

Hypoxia-inducible factor (HIF) is a protein that helps cells respond to changes in oxygen levels. It is made up of two parts, HIF-α and HIF-β, and its stability is controlled by enzymes called prolyl hydroxylases (PHDs or HIFP4H). Under normal oxygen conditions, PHDs promote the breakdown of HIF, keeping its activity low. However, when PHDs are inhibited, HIF becomes stable and active, which can help treat conditions like ischemic diseases and oxidative stress-related disorders. Despite the potential, many drugs targeting PHDs have shown side effects, such as liver issues and high blood pressure. To address these challenges, our research focuses on exploring the chemical space of natural products as potential PHD inhibitors. We aim to identify new small-molecule drugs that selectively target PHD isoforms and design their biosynthetic pathways in collaboration with experimental research groups to develop safer and more effective therapies.

Retinol-binding proteins (RBPs) play a critical role in the uptake, transport, and metabolism of dietary retinoids (vitamin A and its derivatives). These proteins are implicated in various diseases and metabolic syndromes, including obesity, cardiovascular diseases, and macular degeneration. Interestingly, recent findings revealed that RBPs also interact with lipids, suggesting their broader role in lipid metabolism and signaling. Despite this, the molecular mechanisms underlying the distinct binding of RBPs to retinoids and non-retinoids remain largely unknown. To address this gap, our research aims to unravel these molecular mechanisms. Building on our findings, we also seek to discover a novel class of RBP antagonists to selectively modulate their activity in disease states.

Understanding complex molecular dynamics often requires navigating free-energy landscapes with multiple metastable states separated by large barriers. These barriers severely limit the timescales accessible to conventional molecular dynamics (MD) simulations, making it difficult to reliably sample rare but functionally important events. Enhanced sampling techniques address this challenge by introducing collective variables (CVs), a bridge between mechanistic insight and mathematical representation. However, identifying good CVs for complex molecular systems remains a major bottleneck, particularly when relying on manual intuition or trial-and-error approaches. To overcome this limitation, we developed gradient descent-based multiclass harmonic linear discriminant analysis (GDHLDA) and Covalent, a supervised machine learning-based protocol for automated CV discovery. These methods combine physical insight with data-driven learning to produce interpretable and discriminative CVs directly from molecular simulation data. Our ongoing research focuses on extending, optimizing, and generalizing these frameworks to improve their accuracy, efficiency, robustness, and applicability across a broad range of chemical and biological systems.