Optimal transport methods for analyzing highly multiplexed spatial proteomic data

Thesis

Cell microenvironments are complex niches of interacting cells that drive tissue function and disease response. In recent years, in situ multiplexed imaging has en- abled the multi-molecular profiling of single cells while preserving their spatial con- text. Although these new technologies capture cell phenotypes and spatial organization at unprecedented resolution, the technology has many trade-offs, and the analysis of the rich spatial information they provide remains a challenge. In this thesis, we develop novel computational methods tailored for multiplex imaging data leverag- ing principles from Optimal Transport (OT), a branch of statistics for comparing probability distributions.

In the first part, we address the challenge of integrating spatial profiles from different tissues, a task essential to enhancing the information obtained from multiplex experiments. We introduce a novel algorithm for optimizing tissue alignment derived from partial OT, and additionally propose an unbiased statistical estimator for OT distances that scales to massive tissues at a fraction of the cost. We benchmark our method against state-of-the-art algorithms, and demonstrate our method’s improved robustness to rips, tears, and other challenging spatial distortions induced by the multiplexing process.

In the second part, we consider the task of quantitatively characterizing cell microenvironments. We introduce a novel framework that captures both microenvironment cell composition and spatial organization using graph OT, and demonstrate its applications in identifying cell-cell interactions, spatial domains, and disease relevant microenvironments on simulated and real datasets. On pathologist annotated mul- tiplexed lung tissues, we demonstrate that our methods recover microenvironments directly linked to known cell interactions and functions. On multiplexed head neck tumors, we identify microenvironments that are highly predictive of patient outcome. We benchmark our method against two state-of-the-art competitors and various base- lines, and show that our method outperforms all comparisons on each dataset.

Lastly, we present a comprehensive case study that showcases the practical appli- cation of both our methods for spatial integration and microenvironment characterization on a novel dataset of multiplexed mouse colorectal cancer tissues. We align massive tissues with hundreds of thousands of cells, and identify distinct, condition- specific microenvironment subtypes that correspond to unique cell enrichment profiles and spatial domains. Together, our methods provide a powerful approach to mapping the complex landscape of the tumor microenvironment in colorectal cancer.

Authors: Boen, J

• DOI: 10.5287/ora-bpx0m7pkj
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: spatial biology; optimal transport
• Subjects: Mathematics; Oncology; Statistics