dupRadar Outputs

What is dupRadar?

In DNA sequencing, PCR duplicates are straightforward to identify and remove: reads with identical alignment coordinates are likely amplification artifacts. In RNA-seq, the situation is more nuanced.

Highly expressed genes produce many RNA fragments. When sequencing depth is sufficient, some of these independent fragments will share the same start and end positions purely by chance. These are natural duplicates — biologically real, not PCR artifacts. Removing them would distort expression estimates.

dupRadar addresses this by analyzing duplication rates in the context of expression level. Rather than applying a blanket duplicate removal, it models the expected relationship between how highly a gene is expressed and what fraction of its reads are marked as duplicates.

The key insight: at low expression levels, duplicates are almost certainly PCR artifacts. At high expression levels, a baseline level of duplication is expected and biologically meaningful.

Why it matters

Duplicate rate analysis helps answer critical quality questions about an RNA-seq library:

• Was the library over-amplified? If low-expression genes show high duplication rates, too many PCR cycles were used.
• Is the library complex enough? Low-complexity libraries produce high duplication across all expression levels.
• Can I trust the expression estimates? If duplication follows the expected expression-dependent pattern, the data is likely reliable.

This information is essential for experiment QC, troubleshooting library preparation protocols, and deciding whether to include a sample in downstream analysis.

The logistic regression model

dupRadar fits a logistic regression to the relationship between expression level and duplication rate. RustQC implements this same model using iteratively reweighted least squares (IRLS).

The model:

duplication_rate = 1 / (1 + exp(-(intercept + slope * x)))

where x is log10(reads per kilobase).

Intercept: Controls the baseline duplication rate at zero expression.

• Values close to 0 (or negative) indicate good library quality — genes with few reads have few duplicates.
• Values above 0.5 indicate significant PCR over-amplification — even lowly expressed genes have high duplication rates.

Slope: Controls how steeply duplication rises with expression.

• Positive values mean duplication increases with expression, which is the expected biological pattern.
• Very high slope values with a low intercept indicate a well-prepared library where duplication is driven by expression level rather than PCR artifacts.
• A low slope combined with a high intercept suggests uniform over-amplification.

Intercept	Slope	Interpretation
Low (~0)	Moderate-high	Good library. Duplicates are expression-dependent.
High (>0.5)	Low	Over-amplified. Duplicates everywhere.
High (>0.5)	High	Over-amplified, but expression signal still visible.
Low (~0)	Very low	Good complexity, very little duplication at any level.

Output files

All dupRadar output files use the BAM file stem as a prefix (e.g., sample.bam produces sample_dupMatrix.txt) and are written to a dupradar/ subdirectory under the output directory. Use --flat-output to write all files directly to the output directory instead. Each output can be individually enabled or disabled via the configuration file.

• Directorydupradar/

  • sample_dupMatrix.txt Duplication matrix (core data table)
  • sample_intercept_slope.txt Fitted model intercept and slope
  • sample_duprateExpDens.png Density scatter plot
  • sample_duprateExpDens.svg
  • sample_duprateExpBoxplot.png Duplication rate boxplot
  • sample_duprateExpBoxplot.svg
  • sample_expressionHist.png Expression histogram
  • sample_expressionHist.svg
  • sample_intercept_slope_mqc.tsv MultiQC general statistics
  • sample_duprateExpDensCurve_mqc.txt MultiQC fitted curve data

Duplication matrix

File: <sample>_dupMatrix.txt

A tab-separated file with one row per gene and 14 columns. This is the core data table from which all other dupRadar outputs are derived.

Column	Description
ID	Gene identifier from the GTF
geneLength	Gene length in base pairs (non-overlapping exon bases)
allCountsMulti	Total read count including multi-mappers
filteredCountsMulti	Read count excluding duplicates, including multi-mappers
dupRateMulti	Duplication rate with multi-mappers: (all - filtered) / all
dupsPerIdMulti	Number of duplicate reads (multi-mapper inclusive)
RPKMulti	Reads per kilobase (multi-mapper inclusive)
RPKMMulti	Reads per kilobase per million mapped reads (multi-mapper inclusive)
allCounts	Total read count (unique mappers only)
filteredCounts	Read count excluding duplicates (unique mappers only)
dupRate	Duplication rate (unique mappers only)
dupsPerIdUnique	Number of duplicate reads (unique mappers only)
RPK	Reads per kilobase (unique mappers only)
RPKM	Reads per kilobase per million mapped reads (unique mappers only)

Genes with zero reads have a duplication rate of 0 and RPK/RPKM values of 0. The format is identical to R dupRadar’s _dupMatrix.txt output.

Intercept and slope

File: <sample>_intercept_slope.txt

A two-line text file containing the intercept and slope of the fitted logistic regression model:

intercept  0.03186
slope  1.60189

These parameters define the fitted curve: y = 1 / (1 + exp(-(intercept + slope * x))), where x is log10(reads/kbp) and y is the duplication rate.

Plot files

RustQC generates three diagnostic plots in both PNG and SVG formats. See the Interpreting plots section below for how to read each plot.

Density scatter plot

<sample>_duprateExpDens.{png,svg}

Duplication rate boxplot

<sample>_duprateExpBoxplot.{png,svg}

Expression histogram

<sample>_expressionHist.{png,svg}

MultiQC files

• <sample>_intercept_slope_mqc.tsv — Intercept and slope values in MultiQC general statistics format.
• <sample>_duprateExpDensCurve_mqc.txt — Fitted duplication rate curve data points for MultiQC line plot visualization.

These files are designed for integration with MultiQC, enabling dupRadar results to be included in multi-sample QC reports.

Interpreting plots

RustQC generates the same three visualizations as the original dupRadar R package and follows the same conventions. All plots are produced in both PNG and SVG formats.

Density scatter plot

File: <sample>_duprateExpDens.png / .svg

This is the primary diagnostic plot, showing the relationship between gene expression level and duplication rate.

• X-axis: Expression level as log10(reads per kilobase). Each point is one gene.
• Y-axis: Duplication rate, ranging from 0 (no duplicates) to 1 (all reads are duplicates).
• Color: Point density — brighter colors indicate more genes at that position. This helps reveal where the majority of genes fall, even when thousands of points overlap.
• Red curve: The fitted logistic regression model.

Expected patterns in a well-prepared library:

• Genes with low expression cluster at the bottom-left with duplication rates near zero. At low sequencing depth, it is unlikely for independent fragments to share identical start/end positions.
• Genes with high expression have moderately higher duplication rates. This is expected — highly expressed genes produce many fragments, and some will share positions by chance alone.
• The fitted curve rises gradually from left to right, staying below 0.5 for most of the expression range.

Duplication rate boxplot

File: <sample>_duprateExpBoxplot.png / .svg

Shows the distribution of duplication rates across expression-level bins, providing a complementary view to the density scatter plot.

• X-axis: Expression bins (log10 reads per kilobase).
• Y-axis: Duplication rate.
• Boxes: The interquartile range (25th to 75th percentile) of duplication rates for genes within each bin.
• Horizontal line in each box: Median duplication rate for that bin.
• Whiskers: Extend to the most extreme points within 1.5 times the IQR.

Boxes should be near zero at low expression and gradually rise at higher expression levels. Very wide boxes at low expression suggest inconsistent library preparation. The median line within each box should track the general trend seen in the density scatter plot. This plot helps determine whether high duplication rates are expression-dependent (expected) or uniform across all expression levels (problematic).

Expression histogram

File: <sample>_expressionHist.png / .svg

A histogram of gene expression levels across all genes in the annotation.

• X-axis: Expression level as log10(reads per kilobase).
• Y-axis: Number of genes.

A bimodal distribution is common: one peak near zero (unexpressed or very lowly expressed genes) and another peak at moderate expression levels. The position and height of the peaks indicate the dynamic range of the experiment. A narrow distribution shifted to the left suggests low overall sequencing depth or poor library complexity.

How to spot problems

Very high duplication at low expression

If genes with few reads already show high duplication rates, this indicates technical PCR duplication rather than biological signal. The density scatter plot will show points clustered in the upper-left region, and the fitted curve will have a high intercept.

Flat duplication curve

If the duplication rate is uniformly high across all expression levels (a flat, high curve), this suggests a severe PCR amplification issue — the library has very low complexity and most reads are duplicates regardless of gene expression.

Very low duplication everywhere

Duplication rates near zero across all expression levels, including highly expressed genes, may indicate that duplicates were not properly marked in the BAM file. Verify that a duplicate-marking tool was run before RustQC.

Normal vs. problematic libraries

Indicator	Good library	Problematic library
Intercept	Close to 0	Above 0.5
Low-expression genes	Near-zero dup rate	Elevated dup rate
Curve shape	Gradual rise	Flat and high, or steep early rise
Boxplot spread	Tight boxes at low expression	Wide boxes everywhere

Configuring outputs

Use a YAML configuration file to control which dupRadar outputs are generated. All outputs are enabled by default. Set any to false to skip generation:

dupradar:
  dup_matrix: true
  intercept_slope: true
  density_scatter_plot: true
  boxplot: true
  expression_histogram: true
  multiqc_intercept: true
  multiqc_curve: true

References

• dupRadar: Sayols S, Scherzinger D, Klein H. dupRadar: a Bioconductor package for the assessment of PCR artifacts in RNA-Seq data. BMC Bioinformatics. 2016;17(1):428. Bioconductor page
• featureCounts: Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics. 2014;30(7):923-930. Subread/featureCounts