Welcome to the Atlas of Neutrophil Biology

Welcome to v0.3.beta! This release focuses on enhanced user experience and improved dataset management.

🎨 User Interface Improvements

• Modernized homepage with cleaner, more organized layout
• Enhanced dataset selection with visual previews and better filtering
• Improved navigation and visual hierarchy throughout the app

📊 Dataset & Analysis Enhancements

• Major database expansion with additional high-quality neutrophil datasets
• Expanded functional gene signature scoring options
• Better marker detection with user-defined thresholds
• Improved plot resizing and visualization responsiveness
• Enhanced data integration and cross-study compatibility

📋 Documentation & Support

• Added comprehensive changelog with version history
• Enhanced documentation pages with better organization
• Improved user guidance and onboarding experience

📋 See full details: View Complete Changelog for complete version history and technical details.

1. Load dataset from Atlas Collection

2. Subset Cell Populations

3. Re-calculate UMAP

4. Choose analysis

PCA (Principal Component Analysis) is a statistical method used to reduce the dimensionality of data while preserving as much variability (information) as possible.

In simpler terms, PCA:

- Takes data with many features (e.g., thousands of genes)

- Finds new summary axes (called principal components, or PCs)

- Each PC captures as much variation in the data as possible

And you can plot or analyze data using just a few of these PCs (often 2 or 3)

- To visualize high-dimensional data (e.g., scRNA-seq) in 2D or 3D

- To denoise data and focus on the main patterns

- As a preprocessing step for clustering or other dimensionality reduction (e.g., UMAP)

- Principal Components (PCs): New axes formed from linear combinations of original features (e.g., genes). PC1 captures the most variance, PC2 the second most, and so on.

- Scores: The projection of your samples (e.g., cells) on the PCs. Used for visualization.

- SVD: Singular Value Decomposition, a mathematical technique used to decompose a matrix into a product of three matrices: U, Σ, and V. It's a fundamental tool in PCA.

- Seed: A random number used to initialize the PCA calculation. Helps ensure reproducibility of results.

In single-cell data, batch effects can arise due to technical variation across different samples, experiments, or sequencing runs — not true biological differences. Batch correction is the process of adjusting for these effects to better compare cells across conditions.

Examples of batch effects:

- Different donors or patients

- Time points or replicates

- Library preparation batches

- Species or sequencing platforms

Neutrophil Canvas gives you full control over batch correction.

You decide:

- Whether to apply correction or not

- Which metadata variable(s) to correct out (e.g., patient ID, species, tissue)

This flexibility allows you to:

- Correct for known batch effects

- Explore the data without correction

- Compare the effect of different correction methods

Batch correction removes variation explained by the selected factor — this may include true biological signal if misused.

If your goal is to compare disease vs control, do not correct for disease status.

If your goal is to compare across patients, correcting for patient might make sense.

UMAP (Uniform Manifold Approximation and Projection) is a nonlinear dimensionality reduction method used to create 2D or 3D plots from high-dimensional data like single-cell gene expression.

It works by finding a low-dimensional embedding (a mathematical representation) of the high-dimensional data that preserves the local structure of the data.

UMAP is particularly useful for visualizing complex, high-dimensional datasets in a more interpretable 2D or 3D space.

- Each point = a single cell

- Cells closer together = more similar gene expression profiles

- Clusters = groups of similar cells (often representing a state or type)

- UMAP preserves both local structure (neighbors) and global relationships

This makes UMAP especially useful for:

- Identifying cell subpopulations

- Comparing across conditions or treatments

- Finding activation gradients or rare cell types

In Neutrophil Canvas, UMAP is automatically computed based on PCA.

Besides Coloring cells by metadata, gene expression, or pathway scores

You also have access to:

- Interactive select tools

- Download UMAP embedding matrix

- UMAP uses nearest-neighbor graphs built from PCA space.

- The layout is stochastic — repeated runs may look slightly different unless the seed is fixed.

- UMAP is not quantitative — distances on the plot should not be interpreted as exact differences.

- UMAP is a visualization tool — not a clustering or scoring method. Use it to explore the data visually, but interpret clusters using metadata and gene signatures, not UMAP shape alone.

Clustering is the process of grouping cells into clusters based on their similarity in gene expression profiles.

This helps identify distinct cell populations or states that share similar characteristics.

In Neutrophil Canvas, clustering is performed automatically based on the PCA-reduced space (with or without batch correction, depending on your setting).

You can:

- Choose which resolution to use (higher = more clusters)

- View clusters on UMAP or directly use them as filters for population comparisons

Clustering is unsupervised and can be sensitive to parameter changes (e.g., resolution, number of PCs)

Not every cluster corresponds to a biologically meaningful group — always validate with known markers or metadata.

The App performs differential signature scoring to highlight genes or pathways that distinguish one cell group from others. This is similar to differential gene expression, but optimized for single-cell resolution.

In bulk RNA-seq, methods like DESeq2, edgeR, or limma are used to account for count distributions and replicates. But in single-cell RNA-seq, the unit of observation is the individual cell, and variance from group size or sparsity becomes more prominent.

To reduce bias from group size imbalance and scaling effects, we provide three metrics that are better suited for single-cell data:

1. Cohen's d in log-fold change (logFC.cohen): similar to t-test, but it ignores unwanted the effect from highly variable group sizes.

2. AUC: Area Under ROC Curve: related to Wilcoxon rank-sum test or Mann–Whitney U test. It is scale-invariant and less sensitive to the variance of the group when compared to Cohen's d.

3. Cohen's d on only expressed cells (LogFC.detected): algorithm is the same with the previous test, but performs only on cells with at least 1 count of the gene. Might perform better when strictly binary design was given.

The App uses scoreMarkers function from Bioconductor scran package.

How it works inside is summarizing every pair-wise comparison for each group given in the choice.

Let us say, we have the following groups: A, B, C in the dataset.

Default choice will calculate: 1) (B+C)/2 vs A; 2) (A+C)/2 vs B; 3) (A+B)/2 vs C.

Centralized plot helps you to focus on one group, picked the most positive ranked markers and the most negative ranked markers.

But if you are more interested in especially A vs B, then selecting only A and B in the calculation part is necessary.

You may update your new clustering results (e.g. more biologically meaningful clusters) as the group for comparison.

Collecting questions...

No known issues.

- Developed by Dr. rer. nat. Fengjun Zhang

Email: fzhang@uni-muenster.de

UniMS mattermost: @fzhang

X.com: @fengjun_zhang