Professional Coding Project - Example #1

Because most of the code I've written remains proprietary, I created this page to showcase my experience indirectly via the output data visualizations.

Presented here: Data visualizations from a flow cytometry analysis pipeline written in R

Additional demonstration: Cut-and-Run Seq Pipeline

Flow Cytometry Analysis (R)

Technical Overview:

• Investigation of chimeric antigen receptor NK cells (i.e. CAR-NK cells)
• Purpose: analyze the efficiency of CAR-NK cell creation
• Actual R code is proprietary (sorry)

Library	Function
flowCore	Creates the flowSet data type
flowAI	Automated qc
flowAssist	FlowFrame converter
flowWorkspace	gatingSets, cytoSets
ggcyto	Autoplot function
ggplot2	data visualization
openCyto	Automated culstering
ggpointdensity	Graphing
uwot	UMAP dimensionality reduction
cytofkit2	Clustering
ggridges	Fluorescence density plots
flowStats	DataFrame to flowFrame converter

General Schema:

• QC: Raw fluorescence batch comparison
• QC: Fluorescence normalization (gaussNorm)

Gating:

• Non-debris gating
• Singlet gating
• Live cell gating
• Marker gating (CD45+ > ...)

Step 1: Raw Batch comparison

• Examine all markers at the same time for a broad overview of the dataset
  

Step 2: Fluorescence normalization by warpSet or gaussNorm:

• With multiple samples (each row) it becomes increasingly important to normalize fluorescence distributions prior to gating
  
• In this comparison, gaussNorm performed best
  

Step 3: Large Debris Removal:

• Standard removal of non-cell datapoints
  
• In this example, I didn't remove the smaller debris as later analysis indicated they were actual cells.
  

Step 4: Singlet Gating:

• Standard removal of cell doublets
  

Step 5+: Two-color gating examples:

• CD45+ gating for lymphocytes is a typical next step after singlet gating
• Shown - an example 2-color gating step with gates for multiple quadrants
  
• This is an example of multiple gates applid to the same set of cells
  

Gating with Multiple Samples:

• Example of reproducibility of a 2-color gating strategy across 46 normalized samples
  

Dimensionality Reduction: tSNE (colors by cluster)

• I tried dimensionality reduction with two different algorithms to see which performed better in terms of computation time and distinguishing cell clusters (calculated separately of course). UMAP performed better than tSNS, in accordance with various reports.
  

Dimensionality Reduction: UMAP (colors by cluster)

• UMAP projection of the same set of cells
  

Gene expression investigation (UMAP)

• To help identify/verify groups of cells, I colored the UMAP projections according the the level of various cell surface markers.
  

Cluster Identification via Gene Expression Heat Map

• Clustering was performed using cytofkit2
• Clustered heat maps cah help show the similarities/differences between clusters
  
• This plot is marked by cell surface marker, though those would normally be replaced by the corresponding gene label
  

Schema:

• The cells we were examining were immunologically naive. Accordingly, there were few highly differentiated cells, necessitating a "flat" gating scheme.

Outputs:

• One of the general concepts in completing a FC analysis is to combine dimensionality reduction, clustering, and the users's experimetal design (the chosen cell surface markers) to assign cell identity. The triple combination of allows users to cross-reference the UMAP location with cluster identity. When they match, it provides two independent assessments of the cell grouping, providing assurance that the user-assigned cell identities defined by their cell surface markers are accurate.
• One of the mechanisms to extract insights is to look at gene expression in different cell types. For example one could examine the fraction of T-cells that express CCR5/CXCR3 and CCR3/CCR4 to subset CD4+ T-cells, providing information on the relative Th1 vs Th2 type cells.