Interrogating spatially-resolved biomarkers in the tissue microenvironment with quantitative image analysis on Imaging Mass Cytometry datasets
2:30–3:00 pm PDT
Presented By: Trevor McKee, PhD
Research efforts focused on understanding tumor heterogeneity and interrogating the tumor microenvironment have increased substantially in recent history. Studying these in the tissue context requires the ability to stain for multiple markers simultaneously. The technique of Imaging Mass Cytometry provides the ability to multiplex more than 30 markers simultaneously on tissue sections. However, a major challenge for the effective quantitative analysis of multiplex images arising from Imaging Mass Cytometry data is the availability of robust image analysis algorithms, using image segmentation to break the whole image into subregions representing distinct tissues and single cells. Working closely with biological scientists, we have developed an analytical pipeline to isolate individual cells according to biological domain knowledge, incorporating distinct cellular morphology and biomarker information into analytical segmentation strategies. The goal of this work is to advance the accuracy and robustness of image analysis strategies to permit greater depths of interrogation of rich, highly multiplexed Imaging Mass Cytometry datasets.
Developing an Analysis Pipeline for Mass Cytometry Studies
3:00–3:30 pm PDT
Presented By: El-ad David Amir, PhD
Analysis is the biggest bottleneck in transforming CyTOF® data into meaningful biological insights. The analysis pipeline is a conceptual tool from statistics and data science: Input goes through a series of predetermined and logical processing steps, producing a relevant output product. In this talk I will present the concept of an analysis pipeline and its application to mass cytometry data. A common pipeline involves data cleaning, feature generation and hypothesis testing, with various visualizations layered over all three steps. I will then review suggested methods for some of the components. For example, I will present a taxonomy for clustering algorithms that can help compare the different methods and frame them in the context of your project. Furthermore, I will address common analysis dilemmas, such as: Which clustering algorithm should I pick? How many clusters do I need? How do I deal with potential batch effects? Finally, I will draw attention to some of the pitfalls researchers often miss when analyzing their data. The target audience for this presentation is researchers in industry and academia of all levels (from undergraduate students to directors and PIs) who are working with mass cytometry.
FAUST: A new interpretable machine learning approach for automated gating
3:30–4:00 pm PDT
Presented By: Raphael Gottardo, PhD