Yoonji Lee

Recent advances in protein structure determination and prediction, large-scale structural databases, and artificial intelligence have reshaped structure-based drug discovery. Structure-aware artificial intelligence models integrate molecular representation learning with three-dimensional protein information to model interactions, predict complex structures and binding poses, and generate novel molecules. In this review, we follow this paradigm along a continuum from protein–ligand modeling to the de novo design of biomolecular binders. We first outline the molecular and protein representations that render structures computable, together with the growing collection of structural data resources. We then survey state-of-the-art methods across drug–target interaction prediction, protein–ligand complex modeling and docking, de novo molecular generation, and biomolecule design, examining the convergence of docking, structure prediction, and molecular generation within co-folding and diffusion-based frameworks. Despite these advances, prospective experimental validation remains scarce, and persistent limitations such as biased structural coverage, limited and ambiguous negative supervision, fragmented benchmarking, and insufficient mechanistic interpretability continue to constrain real-world utility. Progress in data quality and supervision design, evaluation rigor, and design-relevant interpretability will be essential to translate methodological innovation into practical impact.

P21-activated kinase 4 (PAK4), a member of the PAK family (PAK1–6), was initially recognized for its role in tumor development. Recently, we discovered PAK4’s involvement in triacylglycerol lipolysis in adipocytes. However, its function in adipogenesis remains unclear. Here we show that PAK4 plays a critical role in adipocyte differentiation. Following adipogenic stimulation, PAK4 protein levels increased. Knockdown of PAK4 in 3T3-L1 preadipocytes or human stromal vascular cells, as well as pharmacological inhibition of PAK4 in 3T3-L1 cells, impaired adipogenesis, as indicated by reduced expression of adipocyte marker genes and decreased lipid accumulation. Mechanistically, PAK4 phosphorylated cyclin-dependent kinase 2 at serine 106, a critical step for CCAAT/enhancer-binding protein β expression during mitotic clonal expansion. Consistent with these findings, preadipocyte-specific Pak4-knockout mice exhibited reduced fat mass and smaller adipocytes. These results reveal PAK4 as a crucial regulator of adipogenesis and, together with its inhibitory role in triacylglycerol lipolysis, further underscore its potential as a therapeutic target for obesity treatment.

The metabolic stability of a drug is a crucial determinant of its pharmacokinetic properties, including clearance, half-life, and oral bioavailability. Accurate predictions of metabolic stability can significantly streamline the drug discovery process. In this study, we present MetaboGNN, an advanced model for predicting liver metabolic stability based on Graph Neural Networks (GNNs) and Graph Contrastive Learning (GCL). Using a high-quality dataset from the 2023 South Korea Data Challenge for Drug Discovery, which comprises 3,498 training molecules and 483 test molecules, we presented molecular structures as graphs to capture the intricate structural relationships that influence metabolic stability. A GCL-driven pretraining step was employed to enhance model generalizability by learning robust, transferable graph-level representations. Notably, incorporating interspecies differences between human liver microsomes (HLM) and mouse liver microsomes (MLM) further improved predictive accuracy, achieving Root Mean Square Error (RMSE) values of 27.91 (HLM) and 27.86 (MLM), both expressed as the percentage of parent compound remaining after a 30-min incubation. Compared to traditional approaches, MetaboGNN demonstrates superior predictive performance and highlights the importance of considering interspecies enzymatic variations. In addition, attention-based analysis identified key molecular fragments associated with metabolic stability, highlighting chemically meaningful structural determinants. These findings establish MetaboGNN as a powerful tool for metabolic stability prediction, supporting more efficient lead optimization processes in drug discovery. This study presents MetaboGNN, a novel predictive framework for liver metabolic stability that integrates Graph Neural Networks with Graph Contrastive Learning. By explicitly incorporating interspecies metabolic differences as a dedicated learning target, MetaboGNN achieved enhanced prediction accuracy. Attention-based analysis also identified molecular fragments that are strongly associated with stabilizing or destabilizing effects, facilitating chemical insights during lead optimization.

An abstract is unavailable. This article is available as a PDF only.

Toxicity, defined as the potential harm a substance can cause to living organisms, requires the implementation of stringent regulatory standards to ensure public safety. These standards involve comprehensive testing frameworks, including hazard identification, dose-response evaluation, exposure assessment, and risk characterization. In drug discovery and development, these processes are often complex, time-consuming, and also resource-intensive. Toxicity-related failures in the later stages of drug development can lead to substantial financial losses, underscoring the need for reliable toxicity prediction during the early discovery phases. The advent of computational approaches has accelerated a shift toward in silico modeling, virtual screening, and, notably, artificial intelligence (AI) to identify potential toxicities earlier in the pipeline. Ongoing advances in databases, algorithms, and computational power have further expanded AI’s role in pharmaceutical research. Today, AI models are capable of predicting wide range of toxicity endpoints, such as hepatotoxicity, cardiotoxicity, nephrotoxicity, neurotoxicity, and genotoxicity, based on diverse molecular representations ranging from traditional descriptors to graph-based methods. This review provides an in-depth examination of AI-driven toxicity prediction, emphasizing its transformative impact on drug discovery and its growing importance in improving safety assessments.

Major depressive disorder is a debilitating condition, with many patients unresponsive to conventional monoaminergic antidepressants. Rapid-acting antidepressants such as ketamine and psilocybin offer promising alternatives, relieving symptoms within hours. Ketamine, an NMDA receptor antagonist, and psilocybin, a serotonergic psychedelic primarily targeting 5-HT_2A receptors, both enhance synaptic plasticity in mood-regulating circuits through distinct mechanisms. This review synthesizes recent clinical and preclinical findings on ketamine and psilocybin, emphasizing their molecular targets, circuit-level effects, and converging downstream pathways. A key shared mechanism involves BDNF–TrkB signaling, which promotes spinogenesis and synaptogenesis critical for sustained antidepressant efficacy. We also discuss 5-HT_2A receptor biased agonism as a potential strategy to dissociate psilocybin’s therapeutic effects from its hallucinogenic actions. By comparing their mechanistic profiles, we identify both overlapping and distinct features that may inform the development of next-generation rapid-acting antidepressants. Understanding how serotonergic, glutamatergic, and neurotrophic systems converge may guide the development of fast-acting, durable, and non-hallucinogenic antidepressants.

Key Points: p21-activated kinase 4 (PAK4) phosphorylates and inactivates GSH peroxidase 3 in the kidney after ischemia-reperfusion. Mice lacking PAK4 or treated with PAK4 proteolysis-targeting chimera were protected from kidney damage caused by ischemia-reperfusion injury. PAK4 expression in kidney tissues post-transplant inversely correlated with kidney function. Background: AKI after ischemia-reperfusion remains a substantial perioperative challenge lacking effective treatment. p21-activated kinase 4 (PAK4), a downstream effector of Rho GTPase, has been explored in hepatic ischemia-reperfusion injury, but its role in renal ischemia-reperfusion is unknown. Methods: Wild-type and proximal tubule–specific Pak4 knockout mice underwent 25 minutes of ischemia followed by 24 hours of reperfusion injury. Primary tubular cells and human kidney-2 cells were exposed to hypoxia-reoxygenation injury to investigate the in vitro effect of PAK4. Selective degradation of PAK4 was employed using proteolysis-targeting chimera (PROTAC) to ameliorate AKI. Results: Post–ischemia-reperfusion, the expression of PAK4 was upregulated through hypoxia-inducible factor 1 α in mouse kidneys. Deletion of PAK4 in proximal tubule cells, but not in myeloid cells, significantly mitigated ischemia-reperfusion–induced AKI, as evidenced by decreased levels of BUN, creatinine, tubular necrosis, apoptosis, macrophage infiltration, and lipid accumulation compared with control mice. Further investigation revealed that PAK4 phosphorylated GSH peroxidase 3 (GPx3) at T47, leading to its proteasomal degradation. In addition, pretreatment of mice with the PAK4 PROTAC preserved GPx3 and enhanced fatty acid β -oxidation, thereby protecting against AKI. In kidney tissues from people with a kidney transplant, elevated levels of PAK4 protein and phosphorylation of GPx3 at T47 were observed. Conclusions: Renal tubular PAK4 contributes to tissue damage during ischemia-reperfusion injury, whereas PAK4 PROTAC mitigates ischemia-reperfusion injury by reducing oxidative stress and promoting fatty acid β -oxidation.