An Lab· Korea UniversityFour perspectives on
one question.
How does the genome shape the brain? We pursue one question from four angles — deep learning, single-cell biology, large-scale sequencing, and integrative multi-omics.
Deep Learning for Noncoding Genome
“98% of the genome is noncoding. Its grammar is the real challenge — and where the signal hides.”
We train large language models and deep neural networks on regulatory sequence to predict how rare and de novo variants in enhancers, promoters, and UTRs perturb gene expression. Applied to autism cohorts, this exposes risk that protein-coding analysis misses.
AI-Driven Virtual Cell
“If we had a working model of a neuron's transcriptional state, we could perturb it in silico — thousands of variants per day.”
Assembling a large-scale multi-omics atlas across developing brain. Training foundation models that accept perturbation (variant, drug, knockdown) and emit cell-state shifts. The goal: a virtual cell for AI drug discovery — predicting variant and drug effects before wet experiments start.
Genetic Architecture of Autism
“Autism genetics has been told largely through European cohorts. We study Korean autism families.”
Long-read whole-genome sequencing of one of the largest Korean autism cohorts. Quad and family-based designs, structural variants, and within-family phenotypic deviation (WFSD) to measure de novo mutation impact. Work with clinicians at Seoul National University Bundang Hospital.
Integrative Multi-omics
“Complex disease lives across omic layers. Genomics alone is a shadow on the wall.”
Integrative analyses spanning genomics, transcriptomics, and proteomics — applied to cancer (non-small cell lung), neurodegeneration (Alzheimer's), and neurodevelopment. Published in Nature Communications, Genome Biology, Genome Medicine.