Skip to main content

Research

My research spans computational method development for spatial transcriptomics, genomics and epigenomics of immune regulation, and tools for making AI agent-driven bioinformatics more accessible and trustworthy.

Cell segmentation quality

Imaging-based spatial platforms like 10x Xenium assign millions of transcripts to individual cells — but every downstream conclusion depends on how well cell boundaries were drawn. I develop CRISP (Co-expression Rejection In Segmentation Purity), an open-source tool in R and Python that flags cells co-expressing mutually exclusive marker genes to quantify segmentation errors without ground-truth annotation.

Using CRISP, we benchmarked five segmentation algorithms — including the 10x segmentation kit, Baysor, Proseg, and Segger — across ten mouse organs with a 5,006-gene Xenium panel, revealing a tissue-dependent tradeoff between transcript capture and cell purity.

Preprint

Rose, J.R., Rose, E.S., Assumpcao, J.A.F., Pathak, H., Peck, H.E., Sasser, L.E., Patel, C.J., Vanover, D. & Santangelo, P.J. CRISP enables comparisons of image-based spatial transcriptomic segmentation quality across ten organs. bioRxiv (2026). doi:10.64898/2026.04.16.718947

Xenium analysis pipelines

Alongside the benchmarking work, I’ve built an open-source toolchain for Xenium experiments: Nextflow pipelines for generating alternative segmentations, Seurat-based workflows for QC through annotation, and utilities for splitting and processing tissue microarrays. These are described on the software page.

Mapping mRNA delivery

Spatial transcriptomics can show not just how much of a delivered mRNA reaches a tissue, but which cells take it up. I map delivery tropism across tissues and species for the lab’s mRNA therapeutics work, including a method that measures immunofluorescence intensity inside Xenium cell boundaries to tie transcript delivery to the protein actually produced.

That analysis underpins a study of an engineered thio-amino polymer that delivers mRNA broadly across lung cell types.

Published

Pramod Kumar, E.K., Jung, Y., Rose, E.S., Joo, J.Y., Rose, J.R., et al. An engineered thio-amino polymer achieves potent, safe, and broad cell type delivery in the lung mapped by spatial transcriptomics. Advanced Materials e75699 (2026). Read the paper

During my PhD I used integrated RNA-seq and ATAC-seq to ask how transcriptional and epigenetic states encode immune cell identity — and what happens when that regulation goes wrong.

Epigenetics of immune memory

I studied how transcriptomic and epigenomic states define human memory T cell subsets. We identified chromatin regions whose accessibility before and after activation (poised accessible regions) distinguishes memory subsets and their activation potential.

Published

Rose, J.R., Akdogan-Ozdilek, B., Rahmberg, A.R., et al. Distinct transcriptomic and epigenomic modalities underpin human memory T cell subsets and their activation potential. Communications Biology 6, 363 (2023). Read the paper

Autoimmunity

I also investigated the dysregulation of B cells in lupus. We found that disease activity is reflected in the transcriptome — and written into the epigenome — of otherwise resting, naïve B cells in individuals with the disease.

Published

Rose, J.R., Ishii, Y., Khosroshahi, A., et al. Identification of a systemic lupus erythematosus disease signature in resting naive B cells. Journal of Clinical Investigation 136, e199528 (2026). Read the paper

AI agents are increasingly capable of running bioinformatic analyses end to end — the hard part is trusting the result. I’m working on Stringency, a framework that puts guardrails around agent-driven analysis: it checks that the analysis plan is grounded in best practices before any code runs, keeps a tamper-proof log of every step the agent takes, and pauses at key decisions until a person signs off. The goal is agent-run analyses that a reviewer — or your future self — can trust, reproduce, and audit.

Stringency on GitHub →


For the full publication list, see my Google Scholar profile.