From FastQ to Figure 1: A Step-by-Step Guide to Publication-Ready Single-Cell Visualization

The Role of Figure 1: Setting the Stage

The first figure of your paper sets the tone for the entire manuscript. It tells the reviewer whether the data is clean, the analysis is rigorous, and the biological conclusions are supported by the evidence. A messy, cluttered, or unreadable Figure 1 raises immediate red flags about the quality of the underlying QC and processing.

Your goal is not just to show data, but to guide the viewer's eye through the logic of your experiment: Here are the cells we captured, here is how they group, and here is what they are.

Step 0: The Prerequisites (FastQ to Count Matrix)

While this guide focuses on the visualization aspect, we must acknowledge the journey from raw data. "Figure 1" is rarely just a UMAP. A complete Figure 1 typically includes:

• Panel A (Study Design): A schematic (often created in BioRender or Illustrator) showing sample collection, processing, and sequencing workflow.
• Panel B (Quality Control): Violin plots showing nFeature_RNA, nCount_RNA, and percent.mt. This proves to the reviewer that your beautiful UMAP isn't just clustered debris or doublets.

Once your raw FastQ files have been aligned and quantified (via CellRanger, STARsolo, or Alevin) and passed QC, you are ready for the fun part.

Step 1: The Foundation (Data Prep)

Great figures start with great data. No amount of ggplot wizardry can fix a UMAP derived from poorly normalized data or uncorrected batch effects.

• Feature Selection: Ensure you are clustering on highly variable genes that drive biological signal, not technical noise (like mitochondrial genes or ribosomal proteins, unless that is your focus).
• Batch Correction: If your UMAP shows clusters separated by "Sample_ID" rather than "Cell_Type," you need to integrate your data (using Harmony, Seurat v5 Integration, or scVI) before visualizing.
• Annotation Hygiene: Remove "doublet" clusters and low-quality "debris" clusters before generating final figures. A "clean" UMAP implies a clean dataset.

Step 2: Mastering the UMAP

The UMAP (Uniform Manifold Approximation and Projection) is the centerpiece. But the default output from Seurat or Scanpy is rarely publication-ready.

The Labeling Problem

Legends with 25 colors are cognitively exhausting. Instead, label clusters directly on the plot.

DimPlot(seurat_obj, reduction = "umap", label = TRUE, label.box = TRUE, repel = TRUE) +
  NoLegend() +
  theme(axis.text = element_blank(), axis.ticks = element_blank()) +
  ggtitle("Cell Type Annotation")

The Over-plotting Problem

When you have >50,000 cells, points start to overlap, hiding density. If you have a massive dataset, consider:

• Rasterization: Use ggrastr to rasterize the points while keeping axes and text as vectors. This keeps file sizes manageable.
• Density Plots: Instead of raw points, use density contours (geom_density_2d) to show where the bulk of cells are located.

Step 3: Evidence of Identity (Dot Plots vs. Heatmaps)

Once you show where the cells are, you must prove what they are. This is usually done with marker genes.

Step 4: Color Theory for Genomics

Your choice of color palette can distort data or make it intuitive.

• Categorical Data (Clusters, Samples): Use distinct, high-contrast palettes. Avoid gradients.
  Tools: ggsci (Nature, Lancet, NEJM palettes), tableau20.
• Continuous Data (Gene Expression): Use "perceptually uniform" palettes where the change in color intensity matches the change in value.
  Good: Viridis, Magma, Plasma.
  Bad: Jet/Rainbow (these introduce artificial boundaries).
• Divergent Data (Up/Down Regulation): Use a palette with a neutral center (white/grey) and two distinct colors.
  Good: Scico "Berlin" or "Vik".

The 5 Most Common Visualization Mistakes Reviewers Hate

1. The "Fruit Salad" UMAP: Using a default rainbow color palette with 50 clusters. It's impossible to distinguish cluster 12 from cluster 13. Fix: Group related clusters into broader "super-clusters" for the main figure.
2. Over-plotting: Plotting 100,000 cells with large dot sizes, creating a solid blob where no density information is visible. Fix: Use smaller point sizes (pt.size = 0.1) or density plots.
3. Unreadable Axes: Tiny font sizes that require a microscope to read. Fix: All text should be legible when the figure is printed on a standard A4 page (usually >8pt font).
4. Missing Legends: Showing expression levels without a scale bar. Is red high or low? Fix: Always include a clear, annotated color bar.
5. "Cherry-picked" Markers: Showing only the one gene that works perfectly on a feature plot. Fix: Use a Dot Plot to show the expression of top 5 markers per cluster to prove specificity.

The Toolkit for Success

While Seurat and Scanpy have built-in plotting functions, taking your figures to the next level often requires dedicated visualization libraries. We've included both R and Python options below.

Figure QC Checklist