Dimensionality Reduction in Cytometry: From Data to Embeddings and Back

The Speaker

David Novak - Independent Bioinformatics Consultant | Burns LSC & Ionic Cytometry

David Novak is a bioinformatician and independent consultant specializing in cytometry, NGS, and other high-dimensional biological data. An alumnus of the Saeys Lab (VIB-UGent), he works hands-on with data while collaborating closely with both biological and computational domain experts. He’s led the development of new dimension-reduction approaches (ViVAE, ViScore) and helped build CyTOF-powered multi-organ models of B- and T-cell development (tviblindi). In collaboration with the Vaccine Research Center (NIH), he’s created a large-scale workflow for analysing human immunophenotype changes associated with age and sex, leveraging a >2000-donor cohort (iidx). His portfolio and ongoing work building robust open-source workflows for single-cell analysis can be found at davnovak.github.io.

Summary

Dimensionality reduction (DR) methods, such as t-SNE and UMAP, are commonplace in cytometry data analysis. With increasing numbers of parameters per dataset, low-dimensional embeddings often function as a quality control, exploration, and general visualisation tool. In this webinar, we’ll discuss the purpose and limitations of DR, different families of algorithms, and ways to incorporate DR into analytical workflows.

Embeddings are useful insofar as they reveal patterns of interest. That includes batch effects, outlier populations, or separations of cells by type and state. DR can effectively work as a generator of hypotheses about our data. Beyond this, some embeddings are amenable to downstream analysis, such as signal normalization, clustering, or trajectory inference. However, any reduction of dimensionality risks introducing artifacts and causing errors down the line. Recent advances in DR, its evaluation, and interactive approaches to visualization can help us steer clear of misinterpretation and ultimately lead us to real discoveries.

Learning Objectives:

The webinar will touch on 5 subtopics pertaining to dimensionality reduction (DR), each of which has practical implications: 

1. Understanding linear vs non-linear DR
2. Distinguishing use cases: validation, hypothesis generation, and downstream data processing
3. Local vs global structure preservation
4. Applying evaluation and quality control measures to DR 5. Incorporating DR effectively into computational workflows 

Who Should Attend:

Wet & dry lab cytometry practitioners interested in computational cytometry. Coding experience is not essential.

Keywords: Dimensionality reduction, evaluation, quality control, exploratory data analysis, visualization

CMLE Credit: 1.0