02. From Tissue Averages to Cell-Level Expression

Bulk RNA sequencing (bulk RNA-seq) measures RNA after combining many cells, whereas single-cell RNA sequencing (scRNA-seq) preserves the cell of origin. Their main difference is not the instrument but what one row of the table represents. One bulk column represents a tissue sample, while one single-cell row represents a captured cell.

Question for this lesson: What becomes visible when a tissue average is replaced by a cell-level matrix?

Suppose six cells produced the following illustrative numbers of detected RNA molecules. A real microscopic-droplet experiment approximates these numbers with unique molecular identifiers (UMIs), unique labels attached to original RNA molecules.

cell	EPCAM	CD3D	FAP
cell_1	12	0	0
cell_2	10	0	0
cell_3	1	9	0
cell_4	0	8	0
cell_5	0	0	14
cell_6	1	0	12

Keeping the rows separates cells high in EPCAM, CD3D, or FAP. Summing them into one bulk measurement retains the presence of FAP RNA but loses which cells produced it.

EPCAM, CD3D, and FAP are example genes that can be high in different cell populations. Their names do not need to be memorized here. The important point is that cells from the same sample can have different values.

One average can hide different mixtures

An average response time of 200 ms can mean every request took 200 ms. It can also mean most took 100 ms while a few took two seconds. Once only the average remains, the two scenes cannot be reconstructed.

A tissue-level expression increase has the same ambiguity.

• The same cell type may express the gene more strongly.
• The cell type that already expresses the gene may have become more abundant.

Single-cell RNA-seq preserves a row for each captured cell. This allows separate questions about cell composition and state changes within the same cell type.

For contrast, inspect the bulk gene-by-tissue-sample matrix in the GSE251845 DEG exercise. Each column in that table already combines signals from many cells.

Inferring cell composition from bulk data is not direct profiling

xCell takes a bulk expression matrix and estimates the strength of gene sets characteristic of immune and stromal cell types. Such a characteristic gene set is called a signature, and inferring cell composition from a mixed signal is called deconvolution. This can extract clues from an existing cohort without a new single-cell experiment.

The output is a cell-type enrichment score, not a table of measured cells. The score describes how strongly a signature appears in a sample and is not a cell proportion. The xCell paper states that scores for different cell types should not be directly compared or interpreted as proportions.

method	observed unit	output	direct interpretation
Bulk RNA-seq	tissue sample	gene by sample expression	average tissue expression
xCell	bulk expression	enrichment score by cell type	relative strength of a signature
Single-cell RNA-seq	captured cell	cell by gene counts	differences among captured cells

Captured cells are not the entire original tissue

Rows in the matrix are cells that survived dissociation, were captured, and produced detectable RNA. Fragile cells may break during enzymatic dissociation, while large or sticky cells may enter the instrument less efficiently.

A matrix containing 30% T cells does not prove that exactly 30% of all cells or tissue volume was T cells. Sample collection, dissociation, capture efficiency, and quality-control (QC) thresholds also affect the observed fraction.

Dissociation also removes spatial information. The matrix does not say which cells were neighbours. Questions about location require spatial transcriptomics or tissue imaging.

The data types in this path

stage	output
Cell isolation and address labelling	a DNA copy of RNA, or cDNA, tagged with cell and original-molecule identifiers
Sequencing and primary processing	gene count matrix by barcode
Quality control	an analysis-ready cell by gene count matrix
Clustering and annotation	similar-cell group IDs and cell-type names
Condition comparison	sample-level expression and composition by cell type

Summary

Bulk RNA-seq measures a tissue average. Single-cell RNA-seq preserves expression rows for captured cells, revealing rare populations and cell-level states. Captured cell fractions, however, are not automatically exact estimates of the original tissue composition.

Next, Giving Cells and RNA Molecules an Address explains how thousands of cells can be pooled for sequencing while retaining the cell and original molecule for every short sequence fragment, or read.

---

Sources

• Droplet-based high-throughput single-cell profiling: Zheng et al., Nature Communications (2017)
• xCell output and interpretation limits: Aran et al., Genome Biology (2017)
• Cell-composition bias from dissociation and preservation: Denisenko et al., Genome Biology (2020)