What Single-Cell RNA-seq Taught Me That Bulk RNA-seq Never Could

When I first began analyzing RNA-seq data, everything looked neat, a count matrix, a sample sheet (metadata), and the differential expression table output. But the more I worked with bulk RNA-seq, the more I realized something was missing. I wasn’t seeing the cells. I was seeing an average, a composite voice from a crowd, not the individual stories within it.

That changed when I ran my first single-cell RNA-seq analysis.

Working with a dataset of NSCLC (non-small cell lung cancer) tumor cells from 10x Genomics, I used Seurat to build an end-to-end pipeline: from quality control and normalization to clustering and marker gene discovery. The result was unlike anything I’d seen before, clusters of transcriptionally distinct cells, each carrying unique molecular signals. Genes like ITGA1, LILRB4, and IGLC2 weren’t just differentially expressed, they were spatially restricted, enriched in specific subpopulations, and biologically meaningful.

I couldn’t help but think: None of this would’ve shown up in bulk.

This article isn’t a tutorial. It’s a reflection on what this project taught me, and why single-cell data doesn’t just offer higher resolution, but a different way of seeing biology altogether.

What is single-cell RNA-seq?

Single-cell RNA sequencing (scRNA-seq) allows us to measure gene expression at the level of individual cells, capturing heterogeneity that bulk methods mask. Unlike traditional RNA-seq, which aggregates RNA from thousands or millions of cells into a single measurement, scRNA-seq tells us which genes are being expressed in which cells, and how those cells differ from one another.

It’s not just more data. It’s a shift in perspective.

Why is it important?

Biological systems, especially tumors, are rarely homogeneous. Cancer progression, immune infiltration, and therapeutic resistance, these processes are driven by cell-to-cell variability. Bulk RNA-seq can tell you that a gene is “upregulated,” but it can’t tell you where or in whom.

The dataset I explored

I analyzed a dataset from 10x Genomics containing ~20,000 dissociated tumor cells from NSCLC (non-small cell lung cancer) patients. I processed the data using the Seurat pipeline in R.

How was this analysis different from bulk RNA-seq?

Bulk RNA-seq is built around the idea of comparisons between samples: tumor vs. normal, treated vs. untreated. You get a matrix with genes as rows, samples as columns, and tools like DESeq2 or edgeR handle the rest. Differential expression is done across biological replicates. Clustering happens at the sample level. Heatmaps show averaged expression per condition.

With single-cell RNA-seq, that framework breaks down.

There are no biological replicates. Each column is now a cell. Instead of summarizing variability, you preserve it. And that changes everything:

• Instead of normalizing for library size across samples, you’re performing log normalization and scaling across thousands of individual cells
• Instead of searching for DEGs between groups, you’re discovering clusters of cells that naturally group based on expression

Bulk vs. scRNA-seq comparison table

Even the visuals reflect this shift: I wasn’t looking at condition-level heatmaps or sample-level PCA plots anymore; I was navigating a UMAP landscape, where clusters emerged organically. I wasn’t reading differential gene tables in isolation; I was tracing markers like LILRB4 or ITGA1 back to specific cell clusters, visualizing exactly where those genes were being expressed in the tumor.

Expression of marker genes across UMAP-identified clusters. Unlike bulk RNA-seq, single-cell analysis reveals where each gene is expressed — showing spatially restricted expression of genes like ITGA1, TRGC2, LILRB4, and FCGR2A across transcriptionally distinct subpopulations.

It made me realize how much complexity bulk RNA-seq averages out, and how much we’re missing when we rely solely on it.

Wrapping Up

Single-cell RNA-seq required me to rethink every part of my analysis, not just how I processed the data, but how I interpreted it. Instead of working with sample-level summaries, I was working with cells as individual units of biology. The questions shifted. The methods shifted. Even the visualizations told a different story.

In bulk RNA-seq, I saw which genes were up or down. In single-cell, I saw where they were expressed, and in whom. That resolution revealed cellular subpopulations I didn’t know existed, and genes whose roles only made sense in the context of specific clusters.

It wasn’t just more complex, it was more honest.

[GitHub repo for this project: *link]*