Abstract

Spatial niche identification tools recover the tissue microenvironments that shape cell function, but existing benchmark studies typically rely on a single accuracy metric or focus on a single tissue type. This thesis benchmarks six niche identification tools, including Seurat, NicheDE, BANKSY, scNiche, Novae, and Quiche, which span composition based, spatial embedding, differential expression, graph fusion, foundational model, and differential abundance approaches. A five dimension evaluation framework was used to evaluate the six tools, which was applied across three resolutions using two 10x Genomics Xenium datasets with contrasting spatial organization, including a HER2+ breast cancer tumor and the mouse brain. While no single tool dominated every dimension, Seurat achieved the strongest overall balance on both tissues. BANKSY, Novae, and Quiche each led on a distinct axis, reference structure alignment, spatial coherence, and compositional accuracy respectively, despite lower aggregate ranks. Tool rankings correlated closely between tissues, indicating that relative performance is not purely tissue specific, and resolution choice materially altered several tool’s comparative standing. These findings support selecting a spatial niche identification tool according to which property, such as reference structure alignment, spatial coherence, or compositional accuracy, a given downstream analysis requires, rather than by aggregate rank alone, and the framework developed here provides a reusable basis for that comparison.

Publication Date

7-2026

Document Type

Thesis

Student Type

Graduate

Degree Name

Bioinformatics (MS)

Department, Program, or Center

Thomas H. Gosnell School of Life Sciences

College

College of Science

Advisor

John Ashton

Advisor/Committee Member

Hannah Aichelman

Advisor/Committee Member

Stefan Schulze

Campus

RIT – Main Campus

Download

Additional files available below

Share

COinS