Hyoungjoon Park — AI for Drug Discovery

About

I'm a Ph.D. candidate in the Integrated M.S./Ph.D. program in Computer Science at Yonsei University, working on AI-driven drug discovery in the Data Engineering Lab under the supervision of Prof. Sanghyun Park.

My research focuses on building and validating robust AI models for protein–ligand interaction analysis and generative molecule design. Specifically, I develop diffusion-based frameworks for structure-based drug design (SBDD) that generate selective 3D molecular structures, as well as physics-informed scoring functions for interpretable binding affinity prediction.

With a dual undergraduate background in Computer Engineering and Healthcare Convergence, I aim to bridge the gap between deep learning theory and real-world drug discovery challenges.

Structure-Based Drug Design Protein-Ligand Interaction Diffusion Models 3D Molecular Generation Binding Affinity Prediction Selective Drug Design Equivariant GNNs

Research

Selective Molecular Generation

Making off-target avoidance an explicit generative objective. Dual-affinity guidance injects attractive on-target and repulsive off-target gradients into the denoising process, steering both atom types and coordinates toward a maximized affinity gap.

Diffusion Guidance Selectivity SBDD

Protein–Ligand Binding Affinity

Estimating affinity without a docked complex. Cross-attention over independently encoded protein and ligand embeddings bypasses 3D spatial alignment, making off-target affinity differentiable at every step of the reverse diffusion trajectory.

Cross-Attention Alignment-Free Docking Scores

3D Molecular Representation

SE(3)-equivariant architectures that keep molecular geometry consistent under rotation and translation, and diffusion formulations that generate atom types, bonds, and interatomic distances jointly rather than in sequence.

SE(3)-Equivariance EGNN Distance Matrix

Education

Current
2025.03 — 2030.02 (expected)

Integrated M.S./Ph.D. in Computer Science

Yonsei University, Sinchon Campus

Data Engineering Lab · Advisor: Prof. Sanghyun Park · AI-driven Drug Discovery
Graduate GPA: 4.2/4.3

Graduated
2019.03 — 2025.02

B.E. in Computer Engineering

Yonsei University, Mirae Campus

Double Major: Healthcare Convergence
Total GPA: 3.92/4.3 · Major GPA: 3.93/4.3

Publications

PAKDD 2026 1st Author Regular Paper Published

TheSelective: Dual Affinity-Guided Diffusion for Selective Molecular Generation

Hyoungjoon Park, Hwanhee Kim, Seungyeon Choi, Seungyong Lee, Yoonju Kim, Sanghyun Park

Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD), LNAI 16598, pp. 16–27, Springer, 2026

Overview of TheSelective: an on-target complex graph, a noisy ligand, and an off-target protein are encoded by an SE(3)-equivariant RefineNet; a dual affinity predictor produces on-target and off-target scores that guide atom coordinate and atom type sampling.

TheSelective encodes the on-target complex, the noisy ligand, and the off-target protein through a shared SE(3)-equivariant backbone, then guides reverse diffusion with an attractive on-target and a repulsive off-target affinity gradient.

DOI Code

Dual-affinity guided diffusion that maximizes the on-target/off-target binding gap without requiring off-target docked structures. Evaluated on CrossDocked2020 against TargetDiff, KGDiff, and BInD.

0.975Selectivity, TM-High
+15.5% vs. KGDiff

3.363Selectivity, TM-Low
+12.2% vs. KGDiff

−9.97On-target Vina, TM-High
best among all methods

−9.95On-target Vina, TM-Low
best among all methods

Full results — CrossDocked2020 (Avg. / Med.)

Set	Model	On-Dock ↓	Off-Dock ↑	Selectivity ↑	QED ↑	SA ↑	Success ↑
TM-High	TargetDiff	−7.583 / −7.577	−7.405 / −7.391	0.178 / 0.165	0.469	0.585	91.0%
	BInD	−7.510 / −7.569	−7.391 / −7.422	0.120 / 0.062	0.505	0.658	88.1%
	KGDiff	−9.290 / −9.309	−8.446 / −8.466	0.844 / 0.727	0.527	0.548	85.6%
	TheSelective	−9.969 / −9.958	−8.994 / −9.001	0.975 / 0.923	0.495	0.534	50.2%
TM-Low	TargetDiff	−7.566 / −7.552	−5.567 / −5.564	1.999 / 1.993	0.467	0.583	91.9%
	BInD	−7.536 / −7.589	−5.608 / −5.643	1.928 / 1.934	0.502	0.654	89.1%
	KGDiff	−9.343 / −9.366	−6.345 / −6.393	2.998 / 2.980	0.528	0.546	85.6%
	TheSelective	−9.954 / −9.909	−6.591 / −6.588	3.363 / 3.259	0.511	0.557	56.9%

Selectivity = Off-Dock − On-Dock (kcal/mol); higher is better. QED / SA reported as Avg.
Success = fraction of molecules validly reconstructed by Open Babel and successfully docked against both proteins. The selectivity gain comes at a cost in reconstruction validity; recovering it is ongoing work.

NeurIPS 2025 Published

Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration

Seungyeon Choi, Hwanhee Kim, Chihyun Park, Dahyeon Lee, Seungyong Lee, Yoonju Kim, Hyoungjoon Park, Sein Kwon, Youngwan Jo, Sanghyun Park

Advances in Neural Information Processing Systems 38 (NeurIPS), pp. 125612–125647, 2025

DOI

IEEE BIBM 2025 Published

TheProtein: Evolutionarily Informed Graph-Surface Protein Representation

Seungyong Lee, Seungyeon Choi, Hwanhee Kim, Hyoungjoon Park, Yoonju Kim, Hyunseo Yang, Sanghyun Park

IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 6929–6935, 2025

DOI

IEEE BIBM 2025 Published

TheProperty: Euclidean 3D Molecular Representations for Explainable ADMET Prediction

Yoonju Kim, Seungyeon Choi, Seungyong Lee, Hwanhee Kim, Hyoungjoon Park, Sanghyun Park

IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 6921–6928, 2025

DOI

IEEE BIBM 2025 Published

Fixing Truncation-Induced Mode Collapse in GFlowNets via Pruning Loss

Hwanhee Kim, Seungyeon Choi, Seungyong Lee, Hyoungjoon Park, Yoonju Kim, Sanghyun Park

IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 7460–7467, 2025

DOI

KCC 2026 Published

QEPPI와 생화학적 특성을 이용한 펩타이드의 약물 디자인 당위성 분류 모델

Tae-il Noh, Seungyeon Choi, Hwanhee Kim, Hyoungjoon Park, Seungyong Lee, Yoonju Kim, Seulgi Baek, Minjae Lee, Sanghyun Park

Korea Computer Congress (KCC), pp. 852–854, 2026

Domestic conference (Korean)

KCC 2025 1st Author Published

거리 정보를 통합한 분자 그래프 생성을 위한 Diffusion 모델

Hyoungjoon Park, Seungyeon Choi, Hwanhee Kim, Seungyong Lee, Yoonju Kim, Sanghyun Park

Korea Computer Congress (KCC), Jeju, Korea, 2025

Domestic conference (Korean)

KCC 2024 Published

CPU 기반 이미지 분류 모델의 추론 가속화 성능 비교

Ki Hoon Kwak, SangPil Cho, HoJun Shin, Hyoungjoon Park, GwangHyeon Yun, Young-Rae Cho

Korea Computer Congress (KCC), Jeju, Korea, 2024

Domestic conference (Korean)

Patents

KR Patent Application Filed

Method and System for Generating Selective Ligand

Sanghyun Park, Hyoungjoon Park

Application No. KR 10-2026-0132757, filed 2026.07.20 · Priority: KR 10-2025-0188859, filed 2025.12.03
Applicant: Yonsei University Industry–Academic Cooperation Foundation

Covers the dual affinity-guided generation method behind TheSelective (PAKDD 2026). Supported by the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (No. RS-2023-00229822).

Academic Service

Current
2026 —

Peer Reviewer

IEEE Journal of Biomedical and Health Informatics (JBHI)

Reviewed 2 manuscripts in 2026.

Experience

Current
2025.03 —

Graduate Researcher

Yonsei University, Sinchon Campus

Research on AI-driven structure-based drug design. Developing diffusion-based frameworks for selective molecular generation, and protein–ligand interaction models for binding affinity prediction.

2024.06 — 2024.07

Short-Term International Research Intern

University of Nevada, Las Vegas (UNLV), USA

Completed a machine learning course and carried out a satellite land-cover image segmentation project.

2024.03 — 2024.06

Undergraduate Research Assistant

Applied Data Science LAB, Yonsei Univ. Mirae Campus

Presented paper reviews at lab seminars and assisted research on drug repositioning.

Projects

2025 — Present

TheSelective: Dual Affinity-Guided Diffusion for Selective Molecular Generation

Dual affinity-guided diffusion for selective 3D molecule generation. On-target attractive + off-target repulsive gradients with a scheduled guidance transition. Selectivity 0.975 (TM-High) / 3.363 (TM-Low) on CrossDocked2020, +15.5% / +12.2% over KGDiff. Published at PAKDD 2026.

Diffusion SBDD Selectivity PyTorch

2025.05 — 2025.11

LAIDD Mentoring Project: Small Molecule Generation & Target Activity Prediction

Small-molecule generation and activity prediction against target proteins. AI drug discovery mentoring program hosted by the Korea Pharmaceutical and Bio-Pharma Manufacturers Association (KPBMA).

LAIDD Mol Gen Activity Prediction KPBMA

2024 — 2025

3D Molecular Graph Generation with Distance Matrix

SDE-based diffusion that jointly generates atom types, bonds, and the 3D distance matrix, with Gaussian-kernel distance features integrated into the graph representation.

SDE GNN QM9 ZINC250k

2024.03 — 2024.06

Drug Repositioning with GCNs

Drug repositioning candidate discovery via GCN link prediction over a drug–disease heterogeneous network, evaluated with 10-fold cross-validation.

GCN Heterogeneous Net 10-Fold CV

Awards & Scholarships

2024

2nd Prize, AI-based Medical Data Analysis Competition

Intel Korea

2024

3rd Prize, Digital Healthcare Startup Idea Awards

Yonsei University

2019 · 2024

Academic Excellence Scholarship (×3)

Yonsei University

2024

SW Major Excellence Scholarship

Yonsei University

2020

1st Prize, Winter Java Training Camp

Yonsei University

2019

1st Prize, Summer C Language Training Camp

Yonsei University

Certifications

2025.11

LAIDD Mentoring Project — Small Molecule Generation & Target Activity Prediction

Korea Pharmaceutical and Bio-Pharma Manufacturers Association (KPBMA)

2024.09

TOEIC 860

ETS

2024.08

CDS Big Data Expert Training Camp with KNIME

Yonsei University

2024.02

Naver Cloud DB Computing Program

NAVER Cloud

2024.01

Healthcare Big Data Analysis with RStudio

Commento

2023.10

TOPCIT 425

IITP

2023.06

AI-based Medical Device Software Training

NIDS

Skills

Deep Learning

PyTorch PyTorch Geometric e3nn Diffusion Models SE(3)-Equivariant GNN Transformer

Drug Discovery

RDKit AutoDock Vina Open Babel CrossDocked2020 QED / SA TM-align

Languages

Python Java C SQL R LaTeX

Tools & Infra

Git Linux Docker W&B DDP Multi-GPU LMDB