RNA-seq Quality Control: What to Check Before Differential Expression

If you are planning a bulk RNA-seq differential expression study, you must perform systematic quality control before you run any statistical test. QC is not optional. It protects against false discoveries, technical artifacts, and wasted resources. This guide is for bench scientists, bioinformatics beginners, and analysts who have raw FASTQ files or aligned counts and need a structured checklist to decide what passes and what needs to be removed. The sequence covers raw reads, alignment, gene counts, sample relationships, and documented exclusion decisions.

At a Glance

Decision Criteria

No universal pass/fail thresholds exist because acceptable quality depends on your organism, tissue, and sequencing platform. However, you should define your own criteria before you run the pipeline. The following ranges are typical starting points used in published training materials from the Galaxy Training Network [3] and Bioconductor documentation [4].

Read quality. For Illumina data, a median Phred score above 28 across the read is expected. If more than 10 percent of reads have a quality score below 20 in any position, consider trimming or discarding that library.

Adapter contamination. After adapter trimming, no more than 1 percent of reads should still contain adapter sequences. MultiQC reports can flag libraries that need additional cleaning.

Mapping rate. For a well annotated genome, at least 70 percent of reads should map uniquely to the reference. Lower rates may indicate sample degradation, contamination, or a poor genome assembly. For non model organisms, a lower rate may be acceptable, but you should document and justify it. See the NCBI Sequence Read Archive [5] for typical mapping rates in public datasets.

Library complexity. High duplication levels (above 60 percent) can signal PCR over amplification or low input RNA. If duplication is high but mapping rate is good, you may still proceed with caution, but you should report the complexity to reviewers.

Sample relationships. In PCA, samples from the same condition should cluster together. If a replicate separates by more than two standard deviations from its group on the first or second principal component, investigate technical causes before excluding it.

Practical Workflow: From FASTQ to Excluded Samples

Follow this sequence for each sample. Document every decision in a version controlled file.

Step 1. Inspect Raw Reads

Run FastQC on every FASTQ file. Collate results with MultiQC. Check the per base sequence quality, per sequence GC content, adapter content, and duplication levels. If adapters are present, trim with Cutadapt or Trimmomatic. Re run FastQC after trimming. Do not proceed to alignment until all raw reads pass your defined thresholds.

Step 2. Align or Pseudoalign

Choose a splice aware aligner (STAR or HISAT2) or a lightweight quantification tool (Salmon or kallisto). Map reads to the reference genome or transcriptome. After alignment, collect mapping statistics: total mapped, uniquely mapped, multi mapped, and unmapped. Also record the percentage of reads assigned to ribosomal RNA and mitochondrial RNA. High rRNA (above 10 percent) suggests poor polyA selection. High mtRNA (above 20 percent) may be biological, as seen in studies of metabolic tissues [6], but you should note it as a potential confounder.

Step 3. Quantify Gene and Transcript Counts

Use featureCounts, htseq count, or the built in counting from pseudoaligners to obtain a count matrix. Generate a per sample summary: total counts, number of genes with at least 10 counts, and library size. Plot library sizes as a barplot. Libraries that are more than two fold smaller than the median may be low quality or have degraded RNA.

Step 4. Assess Sample Level Quality

Perform PCA on the variance stabilized or normalized counts (e.g., using rlog or vst from DESeq2). Color by condition, batch, and any technical variable (flow cell, date of sequencing). Look for clustering that matches your experimental design. If a sample separates from its biological group, examine its raw QC metrics. Pair the suspected outlier with a sample from the same condition to see if the pattern replicates the Galaxy Training Network [3] recommendation to check technical replicates before excluding.

Step 5. Make and Document Exclusion Decisions

Create a table with one row per sample. Columns include sample ID, total reads, unique mapping rate, library size, percent mtRNA, and PCA distance to group centroid. Mark samples that fail more than two of your criteria. Write a short justification for every exclusion. Include this table in your methods section or as a supplementary file.

Common Mistakes

Skipping adapter trimming. Even low levels of adapter contamination can bias quantification in the 3 prime end. Always trim and confirm with a second FastQC run.

Removing samples solely based on mapping rate. A low mapping rate could be driven by a new transcript isoform or a contamination that is actually from the same species. Check the unmapped reads to see if they align to a different organism.

Ignoring batch effects from different sequencing runs. If you sequenced samples on two different flow cells or at different centers, include batch as a covariate in your differential expression model. Use Bioconductor packages like sva or limma to detect and correct batch effects [4].

Using correlation as the only sample relationship metric. Correlation is sensitive to library size. Use PCA or MDS on normalized data instead. The EMBL EBI Training materials [2] emphasize that PCA reveals structure that correlation matrices can miss.

Failing to document exclusion decisions. Without a record, you cannot reproduce your analysis or defend it during peer review. Every exclusion should be justified in a code notebook or a lab note.

Limits and Uncertainty

Quality control cannot fix poorly designed experiments. If your control and treatment groups differ only by batch, no QC threshold will remove that confound. Similarly, QC cannot distinguish biological variation from technical noise in every case. A sample that separates in PCA may be a true biological outlier, not a technical failure. Consider running the analysis with and without the suspected outlier and reporting both results.

Thresholds are guidelines, not laws. For example, a study on heat stress in alfalfa reported low mapping rates due to incomplete genome annotation [9]. The authors documented the issue and used pseudoalignment to improve assignment. Do not blindly apply a 70 percent mapping rate cutoff without understanding your system.

Finally, QC is not a single pass. You may need to iterate: after removing one outlier, re run PCA to see if other samples shift. The process stops when no sample fails your criteria and the sample clusters make biological sense.

Frequently Asked Questions

1. What is a good mapping rate for RNA-seq?
For a well annotated model organism with a reference genome, mapping rates above 75 percent are typical. For non model organisms or degraded RNA, rates as low as 50 percent may be acceptable if you use a de novo assembled transcriptome. Always report the mapping rate and justify the threshold.

2. Should I remove genes with very low counts before differential expression?
Yes. Most differential expression tools like DESeq2 or edgeR recommend prefiltering. A common rule is to keep only genes with at least 10 total counts across all samples. This reduces multiple testing burden and removes noise. However, do not prefilter based on differences between conditions.

3. How can I detect a sample swap in RNA-seq data?
Compare the expression of sex specific genes (e.g., XIST in mammals) or genotype markers. Check that the sample’s genetic ancestry matches the reported metadata. If you have known SNP data, you can verify sample identity with tools like verifyBamID.

4. What should I do if my PCA shows one replicate far from its group?
First, check that the sample is not mislabeled by verifying its mapping statistics and library size. Then examine the raw read QC again. If the sample passes all technical metrics, run the differential expression with and without it. If results change dramatically, there may be a biological reason to exclude it, but you must document the decision.

References and Further Reading

1. NCBI Bookshelf. Free biomedical books covering RNA-seq analysis fundamentals. https://www.ncbi.nlm.nih.gov/books/
2. EMBL EBI Training. Official training on RNA-seq quality control and analysis workflows. https://www.ebi.ac.uk/training/
3. Galaxy Training Network. Open tutorials for RNA-seq QC, alignment, and differential expression. https://training.galaxyproject.org/
4. Bioconductor. Software and documentation for genomic analysis, including QC and batch correction. https://bioconductor.org/
5. NCBI Sequence Read Archive. Public repository to compare your mapping rates with published data. https://www.ncbi.nlm.nih.gov/sra
6. Home based high intensity exercise during neoadjuvant chemotherapy and tumor microenvironment remodeling. Cell Commun Signal. Example of tissue specific mtRNA content in QC. https://pubmed.ncbi.nlm.nih.gov/42432678/
7. Transcriptomics reveals L carnitine to enhance semen quality in Malabari bucks. Reprod Domest Anim. Illustrates QC in livestock transcriptomics. https://pubmed.ncbi.nlm.nih.gov/42429318/
8. Non linear transcriptional responses in Tributyltin toxicity. Environ Pollut. Shows how low mapping rates in non model species are handled. https://pubmed.ncbi.nlm.nih.gov/42425285/
9. Single cell transcriptomics and heat stress in alfalfa. Plant Physiol. Discusses QC challenges with incomplete genome annotation. https://pubmed.ncbi.nlm.nih.gov/42424531/
10. Protocol for minimized bias profiling of liver and adipose tissue in mice. STAR Protoc. Provides a detailed QC protocol for single nucleus RNA-seq that can inform bulk RNA-seq practice. https://pubmed.ncbi.nlm.nih.gov/42418327/

Related Articles

• RNA Sequencing Analysis: From FASTQ Files to Biological Questions
• How to Plan a Bulk RNA-seq Differential Expression Study
• Single-Cell RNA-seq Workflow: A Practical Analysis Roadmap
• Single-Cell RNA-seq Quality Control: Cells, Genes, and Mitochondrial Reads
• Cell Type Annotation in Single-Cell RNA-seq: A Layered Evidence Framework