This article presents a comprehensive, R-based practical workflow for preparing spectral flow cytometry datasets for high-dimensional data analysis. With the increasing complexity and dimensionality of flow cytometry experiments, especially with spectral flow cytometry enabling panels of over 40 colors, traditional manual gating methods are insufficient for comprehensive data exploration and reproducibility. The authors address common pitfalls in handling these complex datasets and provide a step-by-step pipeline covering critical aspects such as quality control, data cleaning, transformation, correction for batch effects, subsampling, clustering, and data integration. Using healthy control data as an example, the workflow aims to equip researchers, even those new to the field, with readily available tools to obtain valid, reproducible results, thereby bridging the gap between advanced cytometry techniques and accessible bioinformatics expertise.
Summarizing "How to Prepare Spectral Flow Cytometry Datasets for High Dimensional Data Analysis: A Practical Workflow"
Introduction to Spectral Flow Cytometry: This technique allows for extensive multicolor panels, enabling the simultaneous investigation of a large number of cellular parameters, with panels exceeding 40 colors already developed.
Challenge of High-Dimensional Data: The increased number of markers in spectral flow cytometry creates high-dimensional datasets that traditional 2D manual gating strategies cannot fully comprehend or explore, leading to limitations in discovering unknown cell populations.
Need for Automated Analysis: Unbiased, automated analysis methods (e.g., t-SNE, UMAP, FlowSOM) are crucial for exploring high-dimensional flow cytometry data, identifying unknown cell populations, and comparing them reproducibly. Despite this, many labs lack bioinformatics expertise and stick to traditional manual gating.
Practical Workflow Provided: The article outlines a detailed, R-based pipeline for preparing spectral flow cytometry data for automated analysis, emphasizing usability for researchers with basic R experience. The code is publicly available on GitHub.
Workflow Steps - Quality Control & Pregating:
Manual Data Check: Essential first step to exclude technical artifacts (clogs, doublets, low viability samples) and ensure data quality ("garbage in = garbage out").
Unmixing Optimization: Crucial for spectral data, recommending single-stain controls on the cell type of interest for best unmixing results, with minimal manual compensation adjustments.
Time Gating: Excludes noise due to flow rate inconsistencies.
Excluding Doublets & Dead Cells: Standard practice for clean data; samples with low viability or inconsistent flow rate windows should be excluded.
Pregating Main Population: Manual gating of the population of interest (e.g., CD3+ T cells) to streamline downstream analysis.
Workflow Steps - Data Import and Transformation:
Importing FCS Files: Use the FlowCore package, ensuring no truncation of extreme fluorescence intensity values common in spectral flow cytometry.
Data Transformation (arcsinh): Recommended to stabilize variance between cell populations, improving discrimination and clustering results. Individual cofactors per fluorochrome are advised, typically between 10 and 10,000, chosen to produce clear peaks and negative values around zero.
Workflow Steps - Automatic Quality Control:
Removes anomalies in flow rate, signal acquisition, and outlier events using packages like FlowAI or PeacoQC.
Workflow Steps - Batch Effects Correction:
Addressing Variability: Acknowledges unavoidable differences between sample batches due to handling, staining, and instrument performance.
CytoNorm Method: Recommended for aligning signal variability using control and validation samples, clustering them with FlowSOM to correct for batch-affected markers.
Workflow Steps - Subsampling:
Facilitating Exploration: Advised for large datasets to save computing time and memory.
Methods: Random selection (random downsampling) or density-dependent selection (preferable for rare cell populations). Testing different cell numbers (20,000 to 200,000) is recommended.
Workflow Steps - Clustering and Dimensionality Reduction:
Algorithms:FlowSOM is highlighted as one of the fastest and best clustering algorithms for large datasets, while UMAP and tSNE are commonly used for dimensionality reduction and visualization.
Reproducibility: Setting a seed is crucial for stochastic algorithms to ensure reproducible results.
Parameter Optimization: Advises playing with parameters (e.g., number of neighbors in UMAP, perplexity in tSNE) on a subsampled dataset to find optimal settings.
Post-Clustering Analysis: Use heatmaps of median expression and cell counts per cluster/group for interpretation, and diffcyt for statistical testing.
Advantages of Spectral Flow Cytometry & Workflow:
Enables larger panels and minimizes compensation requirements compared to conventional flow cytometry.
Reduced autofluorescence effects.
Practical aspects like reusing reference controls save time.
Comparable acquisition times to conventional flow cytometry.
The workflow enhances reproducibility and allows for valid comparisons between labs.
Key Takeaways & Future Outlook:
The workflow takes 1-2 days to go through and optimize, with additional time needed for downstream analysis.
Standardizing workflows and potentially panels could significantly improve data integration and statistical power in immunology research.
Encourages sharing flow cytometry data in public repositories to test new integration techniques and foster discoveries.