Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index.

Summarizing and correcting the GC content bias in high-throughput sequencing

Open access
Benjamini, Yuval, Speed, Terence P. · 2012 · Nucleic Acids Research

Abstract

GC content bias describes the dependence between fragment count (read coverage) and GC content found in Illumina sequencing data. This bias can dominate the signal of interest for analyses that focus on measuring fragment abundance within a genome, such as copy number estimation (DNA-seq). The bias is not consistent between samples; and there is no consensus as to the best methods to remove it in a single sample. We analyze regularities in the GC bias patterns, and find a compact description for this unimodal curve family. It is the GC content of the full DNA fragment, not only the sequenced read, that most influences fragment count. This GC effect is unimodal: both GC-rich fragments and AT-rich fragments are underrepresented in the sequencing results. This empirical evidence strengthens the hypothesis that PCR is the most important cause of the GC bias. We propose a model that produces predictions at the base pair level, allowing strand-specific GC-effect correction regardless of the downstream smoothing or binning. These GC modeling considerations can inform other high-throughput sequencing analyses such as ChIP-seq and RNA-seq.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1093/nar/gks001

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1. Single position model estimation. (A) Mappable positions along the genome are randomly sampled (⇓); (B) these positions are stratified by the GC count in the corresponding sliding window (here a = 0, l = 4); (C) the number of fragments (→) with 5′-end (○) in sampled locations are counted; (D) mean fragment rates for each stratum are estimated, taking the ratio between fragment count and positions in the stratum. These form the GC curve (here the curve for a = 75, l = 50 from Figure 4C).
Figure 1. Single position model estimation. (A) Mappable positions along the genome are randomly sampled (⇓); (B) these positions are stratified by the GC count in the corresponding sliding window (here a = 0, l = 4); (C) the number of fragments (→) with 5′-end (○) in sampled locations are counted; (D) mean fragment rates for each stratum are estimated, taking the ratio between fragment count and positions in the stratum. These form the GC curve (here the curve for a = 75, l = 50 from Figure 4C).CC BY-NC · PubMed Central
Figure 2. GC curves (10 kb bins). Observed fragment counts and loess lines plotted against GC of (A) two libraries from the same normal sample, and (B) the tumor library (red) with its matched normal sample library (blue). Counts and curves of all libraries are scaled to fit median counts of normal library 1. Bins were randomly sampled from chromosome 1, and counts include fragments from both strands.
Figure 2. GC curves (10 kb bins). Observed fragment counts and loess lines plotted against GC of (A) two libraries from the same normal sample, and (B) the tumor library (red) with its matched normal sample library (blue). Counts and curves of all libraries are scaled to fit median counts of normal library 1. Bins were randomly sampled from chromosome 1, and counts include fragments from both strands.CC BY-NC · PubMed Central
Figure 3. Single position models. (A) The top curves represent TV scores for GC windows of different lengths, all beginning at 0 (a = 0). The horizontal bars on the bottom mark the median fragment lengths (and 0.05, 0.95 quantiles). For each library, the strongest GC windows are those that encompass the full fragment. For library 1, we mark the optimal model (W0,180), and show its resulting GC curve on the right panel (B). (We actually show W2,176, removing 2 bp from each side of the fragment.) The GC curve measures the fragment rate given the fraction of GC in the window. Vertical lines (blue) represent 1 SD. For comparison, we plot the distribution of GC (dotted line) in our sample from chromosome 1 (scaled).
Figure 3. Single position models. (A) The top curves represent TV scores for GC windows of different lengths, all beginning at 0 (a = 0). The horizontal bars on the bottom mark the median fragment lengths (and 0.05, 0.95 quantiles). For each library, the strongest GC windows are those that encompass the full fragment. For library 1, we mark the optimal model (W0,180), and show its resulting GC curve on the right panel (B). (We actually show W2,176, removing 2 bp from each side of the fragment.) The GC curve measures the fragment rate given the fraction of GC in the window. Vertical lines (blue) represent 1 SD. For comparison, we plot the distribution of GC (dotted line) in our sample from chromosome 1 (scaled).CC BY-NC · PubMed Central
Figure 4. Different lags. (A) GC curve of the window before the fragment, W−50,50; (B) within the read, W0,50 and (C) in the fragment center, not overlapping the read, W75,50. (D) A plot of TV scores for 50 bp sliding windows (Wa,50). The x-axis marks a, the location of the window 5′-end relative to 5′-end of the fragment. On the bottom, we mark a fragment and its reads in relation to the GC windows from the top panels.
Figure 4. Different lags. (A) GC curve of the window before the fragment, W−50,50; (B) within the read, W0,50 and (C) in the fragment center, not overlapping the read, W75,50. (D) A plot of TV scores for 50 bp sliding windows (Wa,50). The x-axis marks a, the location of the window 5′-end relative to 5′-end of the fragment. On the bottom, we mark a fragment and its reads in relation to the GC windows from the top panels.CC BY-NC · PubMed Central
Figure 5. Fragment rate by length and GC. (A) A heat map describes rates for each (GC, length) pair. Each dotted line represents a single length. In (B), GC curves for fragments of specific lengths are drawn [corresponding to the dotted lines in (A)]. Blue / dark curves represent shorter fragments than red / bright. Here x-axis is the fraction of GC. All fragment length models here have a margin of 2 from both fragment ends (a = 2, m = 2).
Figure 5. Fragment rate by length and GC. (A) A heat map describes rates for each (GC, length) pair. Each dotted line represents a single length. In (B), GC curves for fragments of specific lengths are drawn [corresponding to the dotted lines in (A)]. Blue / dark curves represent shorter fragments than red / bright. Here x-axis is the fraction of GC. All fragment length models here have a margin of 2 from both fragment ends (a = 2, m = 2).CC BY-NC · PubMed Central
Figure 6. Fragmentation effect. Left: relative abundance of nucleotides at fixed positions relative to fragment 5′-end. A horizontal dotted line marks the relative abundance of the base at mappable positions. Right: fragment rates when stratifying by the dinucleotide (−1,0). Dinucleotide counts overlapping the fragment 5′-end.
Figure 6. Fragmentation effect. Left: relative abundance of nucleotides at fixed positions relative to fragment 5′-end. A horizontal dotted line marks the relative abundance of the base at mappable positions. Right: fragment rates when stratifying by the dinucleotide (−1,0). Dinucleotide counts overlapping the fragment 5′-end.CC BY-NC · PubMed Central
Figure 7. Aggregation of single location estimates. (A–C) Estimates based on the fragment GC curve (black) trace similar paths as loess (cyan) estimated on observed counts (blue) on multiple scales. (D–F) Estimates based on alternative models compared with observed counts on 1 kb bins. (D) Read model, predictions based on GC of the 5′ read only (W0,75); (E) Two-ended model uses GC (30 bp) from both ends of the fragments; (F) Fragmentation model based on location-specific composition around the 5′-end. See Supplementary methods for details on how models for (E) and (F) were defined and estimated.
Figure 7. Aggregation of single location estimates. (A–C) Estimates based on the fragment GC curve (black) trace similar paths as loess (cyan) estimated on observed counts (blue) on multiple scales. (D–F) Estimates based on alternative models compared with observed counts on 1 kb bins. (D) Read model, predictions based on GC of the 5′ read only (W0,75); (E) Two-ended model uses GC (30 bp) from both ends of the fragments; (F) Fragmentation model based on location-specific composition around the 5′-end. See Supplementary methods for details on how models for (E) and (F) were defined and estimated.CC BY-NC · PubMed Central
Figure 8. Corrected counts of normal sample. (A) Counts in 1 kb bins not corrected for GC (top), corrected by loess (center) and corrected by fragment model (bottom), positions 58 000–62 000 kb of chromosome 1 (chr1). (B) Histograms of the corrected counts (random sample of 1 kb bins in chr1). Each point represents counts from both libraries (forward strand).
Figure 8. Corrected counts of normal sample. (A) Counts in 1 kb bins not corrected for GC (top), corrected by loess (center) and corrected by fragment model (bottom), positions 58 000–62 000 kb of chromosome 1 (chr1). (B) Histograms of the corrected counts (random sample of 1 kb bins in chr1). Each point represents counts from both libraries (forward strand).CC BY-NC · PubMed Central
Figure 9. CN gain from tumor sample. Counts and corrected counts at position 29 000 kb on chromosome 2. (A) Unnormalized counts at 1 kb bins (top), corrected by loess (center) and corrected by fragment model (bottom). GC curves estimated on chromosome 1 (which has no large CN changes). (B) Histogram of normalized counts at 28–30 mb (underlined on left plots).
Figure 9. CN gain from tumor sample. Counts and corrected counts at position 29 000 kb on chromosome 2. (A) Unnormalized counts at 1 kb bins (top), corrected by loess (center) and corrected by fragment model (bottom). GC curves estimated on chromosome 1 (which has no large CN changes). (B) Histogram of normalized counts at 28–30 mb (underlined on left plots).CC BY-NC · PubMed Central
Figure 10. Comparison to Poisson variation. 0.1 and 0.9 quantiles of observed counts grouped by estimated rates. Models that predict better will have narrower vertical spreads. Variation around the mean of the fragment model (green), the loess (blue) and mappability (black) are compared to variation around a Poisson (red).
Figure 10. Comparison to Poisson variation. 0.1 and 0.9 quantiles of observed counts grouped by estimated rates. Models that predict better will have narrower vertical spreads. Variation around the mean of the fragment model (green), the loess (blue) and mappability (black) are compared to variation around a Poisson (red).CC BY-NC · PubMed Central
Figure 11. GC plots for Dataset 2. (A) GC effect for 10 kb (chromosome 1). (B) TV scores for GC windows of different lengths with a = 0 (comparable to Figure 3). (C) GC curve at fragment model (W2,500). (D) Observed (blue) and predicted (black) counts against GC for 10 kb bins (chromosome 2).
Figure 11. GC plots for Dataset 2. (A) GC effect for 10 kb (chromosome 1). (B) TV scores for GC windows of different lengths with a = 0 (comparable to Figure 3). (C) GC curve at fragment model (W2,500). (D) Observed (blue) and predicted (black) counts against GC for 10 kb bins (chromosome 2).CC BY-NC · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 21 references

Cited by

See all 913

Metadata

Authors
Benjamini, Y., & Speed, T. P.
Affiliations
Walter and Eliza Hall Institute of Medical Research (AU)
Corresponding author
  • Benjamini, Yuval
  • Speed, Terence P.
Venue
Nucleic Acids Research
Published
2012-02-09
Volume
40
Issue
10
Pages
e72-e72
ISSN
  • 0305-1048 (ISSN-L)
  • 1362-4962
OpenAlex
W2018604728
Cited by
913
References
21
Citations by year
2026: 52 · 2025: 57 · 2024: 69 · 2023: 47 · 2022: 63 · 2021: 84 · 2020: 74 · 2019: 73 · 2018: 83 · 2017: 69 · 2016: 52 · 2015: 60 · 2014: 72 · 2013: 47 · 2012: 10
FWCI
24.27
Type
journal article
Language
en
Open access
Locations
  • Nucleic Acids Research (publishedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • Minerva Access (University of Melbourne) (submittedVersion, cc-by-nc)
  • HighWire Press Open Archive (submittedVersion)
Funders
  • National Science Foundation
  • National Institutes of Health
  • National Cancer Institute (U24CA143799)
  • National Institute of General Medical Sciences (R01GM083084)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Fractal and DNA sequence analysis
Keywords
GC content bias, read coverage, DNA sequencing, copy number estimation, PCR bias, ChIP-seq, RNA-seq
MeSH terms
Base Composition, DNA, Humans, Models, Genetic, Genome, Human, Poisson Distribution, Sequence Analysis, DNA, High-Throughput Nucleotide Sequencing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Summarizing and correcting the GC content bias in high-throughput sequencing. https://app.cochat.ai/papers/10.1093/nar/gks001
The paper
Benjamini, Y., & Speed, T. P. (2012). Summarizing and correcting the GC content bias in high-throughput sequencing. Nucleic Acids Research. https://doi.org/10.1093/nar/gks001

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.