Jayoung Ryu
Postdoctoral Associate · Biological ML Group (PI: Prof. Romain Lopez) · Courant Institute School of Mathematics, Computing, and Data Science · New York University
I am a Postdoctoral Associate in the Biological ML Group (PI: Prof. Romain Lopez) at New York University, working on ML algorithms for genomics and biological data including cross-modality integration and experimental design.
I completed my Ph.D. in Biomedical Informatics at Harvard University, advised by Luca Pinello. My doctoral work focused on developing computational methods for CRISPR screen analysis, single-cell multiomics, and gene regulatory inference using probabilistic graphical models and graph representation learning.
Research
Experimental design for biological data
My current work focuses on experimental design algorithms tailored to biological tasks such as perturbation prediction — developing principled strategies to select informative interventions in biological systems.
Optimal transport for cross-modal perturbation prediction
I have worked on algorithms that align and predict distributional responses across measurement modalities. My labeled Gromov-Wasserstein optimal transport method enables cross-modality matching and prediction of data with group-level labels, and thereby prediction of single-cell perturbation screen outcomes without requiring paired training data.
Probabilistic modeling of CRISPR screens
I develop Bayesian probabilistic models that jointly account for genotypic editing outcomes and phenotypic screen readouts to improve variant effect quantification. My work on CRISPR-BEAN introduced a graphical modeling framework for base editor reporter screens, enabling accurate identification of causal variants for complex traits such as cellular LDL uptake.
Graph representation learning for single-cell multiomics
I design graph neural network-based methods that embed cells and genomic features jointly to uncover regulatory relationships across modalities. SIMBA formulates single-cell data as a heterogeneous graph, enabling simultaneous analysis of cells, genes, chromatin peaks, and sequence motifs in a shared embedding space.
Publications
Active Learning of Conditional Generative Models via the Transport Neural Tangent Kernel Under review
Ryu J., Cho K., & Lopez R.
LDLR variant classification through activity-normalized prime editing screening Circulation 2026 In press
Zhou P., Simon M., et al., Ryu J., et al., & Sherwood R.
Cross-modality Matching and Prediction of Perturbation Responses with Labeled Gromov-Wasserstein Optimal Transport AISTATS 2025
Ryu J., Lopez R., Bunne C., & Regev A.
Joint genotypic and phenotypic outcome modeling improves base editing variant effect quantification Nature Genetics 2024
Ryu J., Barkal S., et al.
Deciphering the impact of genomic variation on function Nature 2024
IGVF Consortium
SIMBA: single-cell embedding along with features Nature Methods 2023
Chen H., Ryu J., Vinyard M. E., Lerer A. & Pinello L.
A new class of constitutively active super-enhancers is associated with the fast recovery of 3D chromatin loops BMC Bioinformatics 2019
Ryu J., Kim H., Yang D., Lee A. J. & Jung I.