top of page

The Beginner’s Guide to Single Cell RNA-Seq Normalization: What It Is and Why It Matters

If you’re new to scRNA-seq, one step you’ll see in every guidebook is single-cell RNA-seq normalization, or log normalization in most cases. Knowing how to do this step helps you follow the pipeline, but truly understanding why it’s done gives you cleaner, more reliable insights.


Whether you’re just starting or need a quick refresher, this beginner-friendly guide covers the essentials of single-cell RNA-seq normalization as well as explains log normalization and other methods.


What is data normalization?


In a broad data analysis context, normalization means adjusting values so they can be fairly compared. Raw data often comes with differences that don’t reflect meaningful signals but instead come from technical or measurement effects. Normalization corrects for these so that comparisons focus on real patterns.


For example, in image analysis, brightness normalization removes differences due to lighting, not the objects themselves. Or in statistics, scaling variables to a common range (like 0–1) ensures no variable dominates just because of larger raw numbers.


What does single cell RNA-seq normalization mean?


Similarly, in single-cell RNA-seq, each cell can have a very different number of sequencing reads (UMIs) that are only due to biological reasons, such as:

  • Cell size/RNA content - Big cells (e.g., neurons) carry more mRNA than small ones (e.g., lymphocytes), leading to more potential UMIs.

  • Cell cycle & activation - S/G2M or activated cells transcribe more, leading to higher depth; quiescent cells transcribe less and hence, lower depth.

  • Gene architecture - some cells express more long poly-A transcripts (easier to capture in 3′ kits) than others.


In single-cell RNA-seq, sequencing depth per cell can vary a lot due to technical reason or biological factor.

Or they are varied also due to technical factors like:

  • Reverse transcription (RT) efficiency - temperature, enzyme activity, inhibitors, or bead quality change how much cDNA you make.

  • Ambient RNA - free-floating RNA can add a small, noisy background - slightly inflating counts in some low-RNA cells, while also making true signal harder to see.


Hence, single-cell RNA-seq normalization refers to adjusting raw gene counts so that cells can be fairly compared to one another. 


Why single-cell RNA-seq normalization?


Now that we understand what single-cell RNA-seq normalization means, the next question is: Why is it necessary in the first place?


In short, the differences in read depth mentioned above are not true biological signals. For example, if one cell simply has more total reads than another, their raw counts aren’t directly comparable - it’s like comparing how much water is in buckets of different sizes.


If we directly compare raw counts, gene X might look like it’s expressed 5× higher in A. But proportionally, 500/50,000 = 1% and 100/10,000 = 1% — the relative expression is actually the same.

From this example, at first glance, Gene X looks 5× higher in Cell A. But when you normalize by the total reads, 500/50,000 = 1% and 100/10,000 = 1% - meaning the relative expression is actually the same.


Another trap would be comparing naturally larger cells. For example, when comparing neurons to small lymphocytes without single-cell RNA-seq normalization, neurons would look like they overexpress everything - but they’re just bigger and have more RNA.


In raw counts, neurons will appear to have more transcripts than lymphocytes. This means even housekeeping genes (e.g., GAPDH) look higher in neurons simply because they carry more total RNA.

The single-cell RNA-seq normalization step helps scale counts so that we can compare them between different cells (e.g., “how many transcripts from gene X are detected per 10,000 transcripts”).


Common single-cell RNA-seq normalization approaches


In single-cell RNA-seq normalization, common methods used include CPM/TPM scaling, log normalization, SCTransform - each with pros and cons depending on your workflow.


After log-normalization, housekeeping genes look similar across neurons and lymphocytes.

  • CPM/TPM Scaling: For CPM (Counts Per Million), raw counts are divided by the total number of reads per cell, then multiplied by 1 million. TPM (Transcripts Per Million) is similar, but first normalizes for gene length first then scales to 1 million. One problem with this is that it doesn't handle differences in variance across cells very well, so for UMI scRNA-seq, these methods still do not stabilize variance across cells.


  • Log normalization: After scaling (e.g., CPM), expression values are log-transformed: log(x+1), as RNA counts are highly skewed because a few genes dominate (like ribosomal or mitochondrial genes). Log normalization compresses high values and spreads out low values to reduce the dominance of highly expressed genes.


  • SCTransform: This is a more advanced method used in Seurat, which is based on a regularized negative binomial model of gene expression (i.e., a statistical model often used for RNA-seq count data).


Above is the essential info about single-cell RNA-seq normalization we think might help you - it’s a great way to kick-start your journey into single-cell RNA-seq as well. It’s easier to follow scRNA-seq analysis once you understand the purpose of each step. The single-cell RNA-seq normalization step, while simple, is still crucial for getting accurate insights from your data. Our C-DIAM Multi-Omics Studio is built to be beginner-friendly, with an intuitive interface that saves you time and reduces confusion around technical steps like normalization. Feel free to give it a try here:




bottom of page