Research
Figures: FAK activator graphical abstract (Comput Biol Chem 2025, © Elsevier) · KMU-11342 (Pharmaceuticals 2026, CC BY 4.0) · ACh-gain landscape, Suppl. Fig. S1 (Cognitive Neurodynamics 2026, open access) · T-maze task and cumulative reward (Sensors 2024, CC BY 4.0) · CASCADE (Bioinformatics 2026, CC BY 4.0) · SNOMED-CT mapping (Med Biol Eng Comput 2026, open access) · symbolic regression vs machine-learning classifiers, ROC (Biomedicines 2026, CC BY 4.0)
1. AI-Accelerated Drug Discovery
We screen and design candidate compounds with virtual screening, AI-based property prediction and molecular dynamics, then validate them with wet-lab collaborators. Recent results include an FAK activator found in silico (Comput Biol Chem 2025) and two kinase inhibitors characterised with Keimyung University: KMU-11342 against colorectal cancer (Pharmaceuticals 2026) and KMU-11361 for rheumatoid arthritis (Inflammation Research 2026). Ongoing targets include the lysosomal ion channel TMEM175 and REM sleep behaviour disorder.
2. Brain Data & Neuropsychiatry
We build interpretable pipelines for neural data across scales. Highlights: predicting a neuron’s genetic marker from its electrophysiology alone (Brain Res Bull 2019), an inhibitory shift of fast-spiking interneurons in human FCD I epilepsy (Nature Communications 2024), and CASCADE, an open-source pipeline that analyses multi-electrode array recordings from five manufacturers including criticality and avalanche statistics (Bioinformatics 2026). We also study immune–neural crosstalk in Alzheimer’s disease (J Neuroimmunol 2025), EEG-based stress detection (IEEE Access 2026), and in-silico epilepsy models in NEURON/NetPyNE.
3. AI-Driven Clinical Decision Support
We prefer models clinicians can read. We mapped free-text EMR diagnoses to SNOMED-CT with a fine-tuned ClinicalBERT (Med Biol Eng Comput 2026) and derived an interpretable risk equation for post-transplant bloodstream infection with symbolic regression (Biomedicines 2026). Earlier work covers endoscopic image classification (Scientific Reports 2021; J Digit Imaging 2023) and an early, widely cited perspective on ChatGPT in medical education (Anat Sci Educ 2024).
4. Hippocampus × Reinforcement Learning
We test whether the hippocampus implements the successor representation and how neuromodulators shape its learning rules (Biosystems 2022; Front Comput Neurosci 2025). Simulation studies examined weight initialisation (Electronics 2023), transfer in noisy T-mazes (Sensors 2024) and noise resilience in 1D/2D environments (Sensors 2025). Most recently, an acetylcholine-modulated predecessor-feature model lets an agent keep exploring after reward (Cognitive Neurodynamics 2026).
“From synapse to silicon, insight travels — and patient care transforms.”