Research areas

We develop machine learning methods that reflect the structure of biological data. The 5 areas below are what 33 publications since 2017 add up to.

Single-cell representation learning

10 papers · 2025–2026

We learn representations of cells from scRNA-seq data so that clustering, cell type annotation, and batch effect correction can be addressed together rather than as separate steps. Recent work pairs a dual-encoder cross-attention architecture with large language models for annotation.

cross-attentiondual encodercell type annotationlarge language models
Publications 10
  • Integrative evidence-knowledge marker selection enhances LLM-based cell type annotation in single-cell RNA-seq analysisBioData Mining2026International journalSCIEIF 7.9
  • scFANCL: Dual contrastive learning with false-negative correction at cell level for single-cell RNA-seq clusteringBMC Genomics2026International journalSCIEIF 3.9
  • Annotation-free phenotype prediction using knowledge-augmented clustering from single-cell RNA sequencing dataBriefings in Bioinformatics2026International journalSCIEIF 7.3
  • CMAgent: 단일세포 RNA-seq 주석을 위한 LLM 기반 세포 유형–마커 관계 큐레이션 에이전트Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2026Domestic conference
  • HetDrugKG: 이종 지식 그래프 기반 단일세포 약물 유도 유전자 발현 변화 예측Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2026Domestic conference
  • scRNA-seq 기반 표현형 예측을 위한 사전 학습 임베딩을 활용한 환자 단위 메타셀 구성Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2026Domestic conference
  • LLM 기반 세포 유형 특이적 마커 유전자 추출을 위한 자동화 접근법Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference
  • scRNA-seq 데이터 축소 기법을 이용한 환자 표현형 예측의 연산 경량화 분석Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference
  • AMIL: Automated cell grouping framework for phenotype prediction with multiple instance learningProceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference
  • GPT-4를 활용한 마커 유전자 기반 단일세포 세포 유형 자동 주석화 모델Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference

Spatial transcriptomics

5 papers · 2025–2026

We work with transcriptomes measured alongside positional information in tissue: predicting expression from histology images, imputing unmeasured spots, and identifying spatial domains.

histologygene imputationspatial domainmultimodal
Publications 5
  • 국소 유전자 맥락과 Spot과 유전자 간 교차 어텐션을 이용한 공간 전사체 표현 학습Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2026Domestic conference
  • 전역·지역 매칭 기반 공간 전사체와 조직 이미지 정렬 자동화 프레임워크Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2026Domestic conference
  • 클라이언트 기반 GeoMx DSP 공간 전사체 데이터 분석 플랫폼Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference
  • Spatial Transcript와 H&E Whole Slide Image의 정렬 자동화 프레임워크Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference
  • DeepSpot-VAE: H&E 이미지 기반 Xenium 유전자 발현 복원 프레임워크Proceedings of the KIISE Conference (한국정보과학회 학술발표논문집)2025Domestic conference

Multi-omics integration for drug response and phenotype

8 papers · 2020–2026

We integrate transcriptome, methylome, and other omics layers to explain how cells respond to drug treatment and how patient phenotypes differ, using autoencoders, tensor decomposition, and graph attention networks.

drug responsegraph attention networkautoencodertensor decompositionbiomarker discovery
Publications 8

Condition-specific regulatory networks

7 papers · 2017–2026

We reconstruct gene regulatory relationships that hold under a specific condition rather than on average — through transcription factor network propagation, kernel canonical correlation analysis, Gaussian process models for time-series data, and literature mining for miRNA-mRNA targets.

network propagationkernel CCAGaussian processtime-seriesmiRNA-mRNAliterature mining
Publications 7

Machine learning methods for biological data

4 papers · 2018–2025

Method development that is not tied to a single omics modality: alignment-free protein family modeling with convolutional networks, hierarchical clustering for arbitrary-shaped data, and algorithms for quantifying genome editing outcomes.

convolutional networksalignment-freehierarchical clusteringgenome editing
Publications 4

R&D projects

4 ongoing · 2 completed
  • Ongoing2026.04–2029.12
    Development of AI-based game content rating technology for efficient rating classification and post-release management
    Korea Creative Content Agency (KOCCA) · Co-investigator
  • Ongoing2024.04–2028.12
    Development of CAR-Treg cell therapeutics for chronic inflammatory diseases through an AI-based target discovery platform
    Korea Health Industry Development Institute (KHIDI) · Co-investigator
  • Ongoing2026.09–2027.08
    Development of an evidence-grounded AI agent for automated cell type–marker relation curation in single-cell transcriptome analysis
    National Research Foundation of Korea (NRF) · PI
  • Ongoing2022.07–2026.12
    IoMT artificial intelligence and NFT interface standard development for the metaverse
    Korea Planning & Evaluation Institute of Industrial Technology (KEIT) · Co-PI
  • Deep reinforcement learning-based combination optimization for identifying patient-specific multi-omics regulatory networks
  • Completed2025.09–2026.08
    Development of a client-side, high-speed and high-precision web-based analysis platform for detecting and quantifying CRISPR-Cas9-induced large DNA deletions
    National Research Foundation of Korea (NRF) · PI
  • Completed2022.09–2025.02
    Graph neural network based multi-omics integrative method for drug response prediction
    National Research Foundation of Korea (NRF) · PI