We are a computational immunology lab at UNC-Chapel Hill, focused on extracting , organizing, and representing information from immune profiling technologies, such as flow cytometry, mass cytometry. Our overall objective is to identify components of the immune system that can be modulated for therapeutic effect.

Screen Shot 2021-02-10 at 12.29.49 PM.png

We are jointly located in Department of Computer Science and Computational Medicine Program at UNC. Our primary focus is in developing new computational methods for single-cell bioinformatics and computational immunology. We are specifically interested in designing new algorithms for extracting signals from high-throughput immune profiling modalities, such as flow and mass cytometry, and recently imaging mass cytometry. scroll down to read about some of our recent projects!

Research

Featuizing and Creating Summaries of the Immune Sytem : We develop methods to encode the rich information about the immune system collected through modern high-throughput immune-profiling techniques, such as CyTOF and imaging CyTOF into a compact, mathematical abstraction. Such abstractions can be used in downstream clinical prediction tasks.

Metaclustering Approach for Automated Cell Population Discovery Across Multiple Samples. from Stanley et al. Nature Communications. 2020. [Paper]

Metaclustering Approach for Automated Cell Population Discovery Across Multiple Samples. from Stanley et al. Nature Communications. 2020. [Paper]

Set Modeling Meets Flow and Mass Cytometry. from Haidong Yi and Natalie Stanley. Proceedings of ACM-BCB 2021. [Paper on BioArXiv] [Published Paper]

Set Modeling Meets Flow and Mass Cytometry. from Haidong Yi and Natalie Stanley. Proceedings of ACM-BCB 2021. [Paper on BioArXiv] [Published Paper]

 

General. We are interested in developing methods to understand flow and mass cyometry data in clinical settings. The algorithms we develop enable unsupervised automated cell-population discovery, extraction of biologically meaningful features, and comprehensive visualization. The ultimate goal of these algorithms is to uncover cell-types and signaling pathways that are implicated in some phenotype of interest. We develop our methods in a scalable way so that they can easily accommodate hundreds of samples with millions of cells.

 

Tracking Cell Populations Identified Automatically. Our recent work enables efficient automated cell-population discovery in hundreds of patient samples without throwing any information away. We can further study these identified cell-populations and identify differences in frequency or function between patient phenotypes. These analyses can be used as a complement to or in place of manual gating.

 

Modeling Cytometry Data as Sets. Our recent work explores modeling flow and mass cytometry data as a collection of sets. A set representation is useful for samples profiled with flow and mass cytometry since 1) a variable number of cells are profiled per sample and 2) the order in which cells are profiled as no biological relevance. To predict patient phenotypes, our objective was therefore to predict a label for each set. This was achieved by applying a deep learning approaches for set classification.