Skip to content

Quality Control in RNA-Seq: MultiQC Report

Quality Matters

Before trusting any downstream analysis — differential expression, pathway analysis, or otherwise — you need to verify that your raw data and processed outputs meet basic quality standards. Poor-quality RNA-seq data can arise from many sources:

  • Degraded RNA at time of library preparation
  • Low sequencing depth
  • Adapter contamination
  • rRNA or DNA contamination
  • PCR over-amplification (duplicates)
  • Biased coverage (e.g., 3' bias in degraded samples)
  • Sample swaps or mislabeling

The nf-core/rnaseq pipeline automates QC at every stage and compiles all results into a single MultiQC report. This section explains each QC tool, what it measures, and what to look for.

The MultiQC Report: Key Points

MultiQC aggregates the output of all QC tools across all samples into a single, interactive HTML report.

Download the html file from the output folder. You will find the report at the following location:

/blue/bioinf_workshop/share/nfcore_rnaseq_files/outputs/multiqc/star_rsem

To use scp to download the report:

# Copy to your local machine and open in a browser
scp username@hpg.rc.ufl.edu:/OUTPUT/star_rsem/multiqc/star_rsem/multiqc_report.html ./
open multiqc_report.html   # macOs

To download it using OnDemand, log in and navigate to the same location and click 'Download' after selecting the html report.

The report includes sections for every tool that ran — FastQC, Trim Galore, STAR, Picard, RSeQC, Qualimap, Preseq, and RSEM. Each section shows per-sample metrics and highlights any samples that fall outside expected ranges.

1. FastQC — Raw Read Quality Assessment

FastQC is the standard first-pass QC tool for sequencing data. It runs on raw FASTQ files and generates a set of diagnostic plots.

Module What to Look For
Per-base sequence quality Phred scores should be >28 across the read. It's normal for quality to drop slightly at the 3' end.
Per-sequence quality scores The distribution should peak at high quality (>30). A low-quality peak indicates a problem.
Per-base sequence content The first \~10 bases often show variation due to random hexamer priming — this is normal. Uniform composition expected after that.
GC content Should follow a smooth, roughly normal distribution centered near the expected GC% for your organism. Unusual peaks or bimodal distributions suggest contamination.
Sequence duplication levels High duplication is expected for RNA-seq (some transcripts are highly expressed). Extremely high duplication (>80%) may indicate over-amplification.
Adapter content Should be near zero after trimming. High adapter content in raw reads is expected and normal before trimming.
Overrepresented sequences rRNA or adapter sequences showing up here suggest contamination.

2. Trim Galore / Cutadapt — Adapter Trimming and Quality Filtering

Trim Galore is a wrapper around the Cutadapt adapter trimmer, with additional quality trimming. It removes adapter sequences and low-quality bases from read ends.

  • Detects and removes Illumina adapter sequences

  • Trims low-quality bases from the 3' end (default Phred \< 20)

  • Removes reads that become too short after trimming (default: \< 20 bp)

  • Runs FastQC on the trimmed reads automatically

  • % reads passing filter: Should be >90% for a good library

  • % bases trimmed: A small percentage (1–10%) is normal. Very high values (>30%) suggest adapter contamination or low quality

  • Reads significantly shorter after trimming may indicate RNA degradation

3. SAMtools — Alignment Statistics

SAMtools is a suite of tools for manipulating SAM/BAM alignment files. In the pipeline, it is used to generate per-sample alignment summary statistics via samtools flagstat and samtools stats.

Metric Meaning What to look for
\% mapped reads Fraction of reads that aligned to the genome >75% (typically 85–95% for good samples)
\% properly paired Both reads in a pair align to the same chromosome in expected orientation >90%
\% singletons One read mapped but its pair did not Should be low (\<5%)
\% reads mapped to different chromosomes Indicative of chimeric or contaminating reads Should be very low

⚠️ A low overall mapping rate (\<70%) can indicate issues such as wrong genome used, high contamination, or very low quality /degraded input RNA.

4. DupRadar - Duplication Rate in Context of Expression Level

DupRadar is a Bioconductor R package that assesses duplication rates in the context of gene expression levels. It goes beyond what Picard's MarkDuplicates offers by asking a more nuanced question: are duplicates arising from highly expressed genes (expected), or are they uniformly high across all genes (problematic)?

  • A healthy library shows low duplication at low expression and high duplication only at high expression
  • A problematic library shows uniformly high duplication across all expression levels
  • Duplication rate vs. expression level: Plots the duplication rate of each gene against its expression level (RPK)

5. Picard — Duplicate Marking

Picard is a Java toolkit from the Broad Institute for processing sequencing data. The MarkDuplicates tool identifies reads that are likely PCR or optical duplicates — identical copies of the same original DNA fragment.

Two reads are considered duplicates if they map to the exact same genomic coordinates. Duplicates are flagged in the BAM file but not removed by default in this pipeline.

Metric Meaning What to look for
\% duplicate reads Fraction of reads flagged as duplicates >30–40% may indicate over-amplification
Estimated library size Estimated number of unique molecules in the library Very small values (\<1M) suggest poor complexity

💡 Some duplication is expected and unavoidable in RNA-seq, especially for highly expressed transcripts. Very high duplication rates combined with low library complexity estimates indicate a problem.

6. Preseq — Library Complexity Estimation

Preseq estimates how many unique reads you would expect to obtain if you sequenced the library to greater depth. This tells you whether your library is saturated (you've already captured most of the diversity) or whether deeper sequencing would yield more information.

  • Expected distinct reads as a function of total reads sequenced
  • A steep, still-rising curve indicates the library has not been exhausted and would benefit from deeper sequencing
  • A curve that flattens early indicates the library is saturated — more sequencing yields diminishing returns

What to look for

  • Samples where the curve rises steeply: good complexity, potentially under-sequenced
  • Samples where the curve plateaus early: low complexity — may indicate heavy duplication or degraded input RNA

7. RSeQC — RNA-Specific Alignment Quality

RSeQC is a Python package specifically designed for assessing the quality of RNA-seq alignments. Unlike SAMtools (which gives generic alignment metrics), RSeQC provides RNA-aware diagnostics.

Key modules used by nf-core/rnaseq:

infer_experiment Infers the strandedness of the library by examining how reads align relative to annotated gene features.

  • Confirms or contradicts the strandedness setting in your sample sheet
  • Important for downstream quantification accuracy

read_distribution Categorizes reads based on which genomic feature they overlap: CDS exon, UTR, intronic, intergenic, etc.

Expected What a Problem Looks Like
Majority of reads in CDS exons and UTRs High intronic reads: unspliced RNA, nuclear contamination, or genomic DNA contamination
Low intergenic reads High intergenic reads: wrong genome or annotation, or heavy contamination

junction_annotation Compares splice junctions in the alignments to known junctions in the annotation.

  • Most junctions should be known (annotated)
  • A high proportion of novel junctions may indicate a poorly annotated genome or contamination

junction_saturation Shows whether you've sequenced deeply enough to detect most splice junctions. A plateauing curve means you've captured most junctions.

inner_distance Estimates the insert size of your paired-end library. Useful for detecting grossly abnormal library preparations.

bam_stat Provides a summary of alignment flags — similar to samtools flagstat but with additional RNA-relevant categories.

8. Qualimap — Coverage and Bias Assessment

Qualimap provides a comprehensive set of coverage-based QC metrics for RNA-seq BAM files.

Metric Meaning
5'/3' bias Measures whether reads are uniformly distributed across transcripts. High 3' bias indicates RNA degradation.
Coverage across transcript A flat line is ideal. A downward slope toward the 5' end indicates degradation.
Reads aligned to genes Should be high — low values suggest annotation or alignment problems.
Junction reads Proportion of reads spanning splice junctions — expected to be high for mRNA.

💡 The 5' to 3' coverage plot is one of the most important diagnostics for RNA quality. If you see a strong 3' bias, it may indicate the RNA was degraded before library preparation, which can bias expression estimates toward 3' transcript ends.

QC Summary: Red Flags to Watch For

Problem Likely QC Signal Possible Cause
Low mapping rate (\<70%) SAMtools flagstat Wrong genome, heavy contamination, very low quality input
High duplication rate (>50%) Picard MarkDuplicates Over-amplified library, low input RNA
Strong 3' bias Qualimap coverage plot RNA degradation
High adapter content FastQC adapter content Sequencing short inserts, trimming failure
Unusual GC distribution FastQC GC content Contamination, PCR bias
Low complexity Preseq Degraded RNA, over-amplification
High intronic reads RSeQC read distribution Genomic DNA contamination, unspliced RNA
Strandedness mismatch RSeQC infer_experiment Incorrect strandedness specified in sample sheet

Using the MultiQC Report for Sample-Level Decisions

Before proceeding to differential expression, use the MultiQC report to:

  • Identify outlier samples — samples that cluster away from others in QC metrics
  • Confirm acceptable mapping rates — flag any sample below 70%
  • Check for batch effects — do samples from the same sequencing run cluster together?
  • Decide whether to exclude any samples — document any exclusions clearly
  • Verify strandedness was correctly detected — check the RSeQC infer_experiment section

💡 It is better to remove a poor-quality sample from the analysis than to include it and risk confounded results. Document all exclusions in your methods.

References