Ranek, J. | Search bioRxiv

bioRxiv2024

Automated classification of cellular expression in multiplexed imaging data with Nimbus

Multiplexed imaging offers a powerful approach to characterize the spatial topography of tissues in both health and disease. To analyze such data, the specific combination of markers that are present in each cell must be enumerated to enable accurate phenotyping, a process that often relies on unsupervised clustering. We constructed the Pan-Multiplex (Pan-M) dataset containing 197 million distinct annotations of marker expression across 15 different cell types. We used Pan-M to create Nimbus, a deep learning model to predict marker positivity from multiplexed image data. Nimbus is a pre-trained model that uses the underlying images to classify marker expression across distinct cell types, from different tissues, acquired using different microscope platforms, without requiring any retraining. We demonstrate that Nimbus predictions capture the underlying staining patterns of the full diversity of markers present in Pan-M. We then show how Nimbus predictions can be integrated with downstream clustering algorithms to robustly identify cell subtypes in image data. We have open-sourced Nimbus and Pan-M to enable community use at https://github.com/angelolab/Nimbus-Inference.

bioinformatics↗

bioRxiv2023

The Interplay Between Sketching and Graph Generation Algorithms in Identifying Biologically Cohesive Cell-Populations in Single-Cell Data

High-throughput single-cell immune profiling technologies, such as mass cytometry (CyTOF) and single-cell RNA sequencing measure the expression of multiple proteins or genes across many individual cells within a profiled sample. As it is often of interest to identify particular clusters or cell-populations driving clinical phenotypes or experimental outcomes, there is a critical need to develop automated bioinformatics approaches that can handle a large number of profiled cells. For analyzing multi-sample single-cell datasets at scale, the datasets are usually encoded as a graph, where nodes represent cells and edges imply significant between-cell similarity. As multi-sample single-cell experiments can readily result in millions of profiled cells, the construction and analysis of a graph becomes computationally prohibitive and often requires reducing the dataset size through downsampling as a pre-processing step. Here, we explore the interplay between sketching, or downsampling approaches, and the way in which the graph is constructed on the sketched data for ultimately identifying biologically-meaningful cell-populations. Our results suggest that combining a principled sketching approach with a simple k-nearest neighbor graph representation of the data can identify meaningful subsets of cells as robustly as, and sometimes better than, more sophisticated graph generation approaches. This reveals that the practical concern of downsampling or sketching a limited number of cells is a more critical pre-processing step than how the graph representation is constructed.

bioinformatics↗