Harness your current Pipeline: How Pixelgen Data Integrates With Your Existing Single-Cell Workflow
Following processing of your raw FASTQ files with our nf/core-pixelator pipeline, you can analyze your data in your favourite coding language, Python or R, and create familiar AnnData or Seurat objects for further downstream analysis.
The analysis workflow follows the same steps as your established single cell omics workflows, and you can easily integrate your preferred visualizations and scanpy/Seurat functions with your PNA data. We also provide in-depth tutorials and analysis templates to help you step by step on the way, including detailed explanations and code so you can analyse the proximity scores of your cells.
Key steps of downstream Proximity Network data analysis
PNA analysis uses familiar data structures, workflows, functions and visualizations from your established single cell analyses. Both pixelator in python and pixelatorR in R support the common single cell omics workflow, including QC, data normalization, dimensionality reduction, integration, and clustering. In addition, PNA allows for deeper analysis of spatial patterns of the panel markers
The starting point for all downstream analysis is the .pxl file produced by pixelator, containing per-cell proximity graphs, the protein abundance count matrix, proximity scores, and metadata. In Python and R, you can load your .pxl files directly into familiar objects such as Seurat or AnnData objects.
02 Count normalization & scaling (standard single-cell workflow)
CLR (centred log-ratio): Each marker count is divided by the geometric mean of all counts in that cell, then log-transformed. Originally developed for CITE-seq, CLR is directly applicable to PNA abundance data.
Scaling: Counts are mean-centered and scaled to unit variance per marker, preventing high-abundance proteins from dominating in dimensionality reduction and clustering.
03 Dimensionality reduction (standard single-cell workflow)
PCA: Run principal component analysis (PCA) to reduce the dimensionality of the data. Select principal components to capture the dominant axes of variation in protein abundance across cell states.
UMAP: Projects the PCA space into 2D, preserving both local cell similarity and global population separation.
04 Data integration (standard single-cell workflow)
Harmony: When combining multiple samples or batches, Harmony iteratively adjusts PCA embeddings to align cells by cell state, reducing the effects of technical differences on clustering and dimensionality reduction. This is recommended whenever samples differ by processing day, operator, or antibody lot. The corrected embedding is used to simplify downstream clustering and visualisation.
Clustering: The protein abundance data is used to identify cells with similar protein abundance profiles and partition them into clusters of cell state.
Cell annotation: Clusters are labelled by inspecting enriched canonical markers, either through visual exploration or differential testing. Our packages also provide functions for flow-like density scatter plots to set expression thresholds and apply gating-based strategies familiar to flow cytometry.
Differential abundance: Once populations are defined, marker abundance can be compared across conditions using standard scanpy/Seurat functions.
06 Protein spatial analysis
This is the step that distinguishes PNA from all other single-cell assays — proximity scores capture not just how much of each protein is present, but where proteins sit relative to each other on the cell surface.
Mean proximity profiles: Cell-type mean proximity scores are visualised as dot heatmaps, where the dot size reflects the fraction of cells in which a given pair is detected and the color indicates the colocalization strength.
Differential proximity: Proximity scores are compared between conditions or cell types using dedicated pixelator/pixelatorR functions,
identifying pairs that shift their spatial relationship. Results are visualised as volcano plots, network plots, and heatmaps.
Cell visualisation: Individual cell graphs are loaded from the .pxl file and visualized in 2D or as a spinning 3D cell, with markers of interest colored to visualize spatial distribution
