ChromCall

Assigning chromatin status to predefined genomic regions from epigenomic profiling data

---

🔍 Overview

ChromCall is an R package for region-based chromatin enrichment analysis of epigenomic profiling data, including ChIP-seq, CUT&RUN, CUT&Tag, and ATAC-seq. It provides a transparent, statistically principled framework to quantify enrichment at predefined genomic regions (e.g. promoters or enhancers), enabling region-matched comparisons across samples and experiments without relying on data-dependent peak boundaries.

---

🚀 Key Features

• Region-centric analysis
  Quantifies chromatin enrichment directly within predefined genomic windows, enabling consistent, region-matched comparisons across samples and experiments.

• Transparent statistical framework
  Employs a Negative Binomial–based background model incorporating:

  • experiment-specific genome-wide background estimation
  • region-specific, control-derived modulation factors (when available)

• Control-aware enrichment testing (optional)
  Supports integration of matched control experiments to account for local background variation. In the absence of a control, ChromCall relies on genome-wide background estimation.

• Multiple complementary metrics
  For each region and experiment, ChromCall reports:

  • FDR-adjusted p-values
  • enrichment score (log₂ observed / expected)
  • z-score
  • significance-based region classification

• Multi-experiment and multi-sample support
  Supports joint analysis of multiple chromatin marks and pairwise comparisons between samples within a unified framework.

• Optional expression integration
  Gene-level expression values mapped to regions using TSS-based annotation can be incorporated to enable integrated chromatin–transcription analyses.

---

🧠 Statistical Model Overview

ChromCall models read counts using a Negative Binomial (NB) distribution, allowing for overdispersion beyond the Poisson assumption arising from technical variability in sequencing depth, library preparation, and local chromatin accessibility, thereby providing a more robust framework for sequencing-based count data.

Background Estimation

For each experiment, a genome-wide background rate ($\lambda_g$) is estimated as the mean read count per non-blacklisted genomic tile:

$$ \lambda_g = \frac{1}{N} \sum_{i=1}^{N} y_i $$

where ($y_i$) denotes the read count in the ith tile and N is the total number of non-blacklisted tiles.
Zero-count tiles are retained by default to avoid upward bias in sparse datasets and to ensure that ($\lambda_g$) reflects global background signal rather than local enrichment.

Control-based Local Modulation

To account for region-specific variability, ChromCall optionally derives a modulation factor from a matched control experiment:

$$ m_i = \max\left(1, \frac{y_i^{(\mathrm{ctrl})}}{\lambda_g^{(\mathrm{ctrl})}}\right) $$

The expected signal for region i in experiment j is then defined as:

$$ \lambda_{t,i}^{(j)} = m_i \times \lambda_g^{(j)} $$

In the absence of a control dataset, $m_i$ defaults to 1, and $\lambda_{t,i}^{(j)}$ is determined solely by genome-wide background estimation.

Statistical Testing and Effect Sizes

ChromCall evaluates region-level enrichment using a one-sided Negative Binomial test:

$$ p_i^{(j)} = P\left(Y \ge y_i^{(j)} \mid Y \sim \mathrm{NB}(\mu = \lambda_{t,i}^{(j)},\ \text{size} = \theta^{(j)})\right) $$

where $\theta^{(j)}$ is a global dispersion parameter estimated per experiment. The dispersion parameter is estimated from genome-wide background tiles, enabling stable estimation by borrowing information across regions in the absence of replicates.

When dispersion is negligible (i.e. $\theta \to \infty$), the model reduces to a Poisson distribution.

Multiple testing correction is applied across all regions using the Benjamini–Hochberg false discovery rate (FDR) procedure.

In addition to significance testing, ChromCall reports complementary effect-size metrics:

• Enrichment score

$$ s_i^{(j)} = \log_2\left(\frac{y_i^{(j)} + \epsilon}{\lambda_{t,i}^{(j)} + \epsilon}\right) $$

• z-score

$$ z_i^{(j)} = \frac{y_i^{(j)} - \lambda_{t,i}^{(j)}}{\sqrt{\lambda_{t,i}^{(j)} + \frac{(\lambda_{t,i}^{(j)})^2}{\theta^{(j)}}}} $$

Together, these metrics provide complementary measures of enrichment strength, effect size, and statistical confidence.

---

🧬 Implementation and Data Structures

ChromCall is implemented in R and builds upon the Bioconductor ecosystem, ensuring interoperability with standard genomic data structures and downstream analysis workflows:

• GRanges for representing genomic intervals
• SummarizedExperiment for storing structured assay outputs and metadata
• GenomicAlignments for importing aligned sequencing reads from BAM files
• GenomeInfoDb and Seqinfo for genome annotation and consistency checks

Each processed sample is returned as a SummarizedExperiment object containing raw region-level read counts together with experiment-level background parameters (bg_mean, bg_size). After statistical testing, additional assays including lambda_t, p_value, p_adj, score, and z_nb are appended.

Pairwise sample comparisons generate region-level Δ enrichment and Δ z-score metrics, enabling direct comparative analysis of chromatin states across biological conditions.

---

📦 Installation

ChromCall is available as a development version on GitHub and can be installed using remotes:

# install.packages("remotes")
remotes::install_github("GliomaGenomics/ChromCall")

---

🧪 Basic Workflow

Build a ChromCall sample

sample <- build_chromcall_sample(
  sample_name   = "sampleA",
  experiments   = list(
    H3K27me3 = "h3k27me3.bam",
    H3K4me3  = "h3k4me3.bam",
    Control  = "control.bam"
  ),
  control_name   = "Control",
  genome_file    = "genome.txt",
  region_file    = "promoters.bed",
  window_size    = 2000,
  blacklist_file = "blacklist.bed",
  expression_file = "expression_tss.bed"
)

Perform region-level enrichment testing

result <- test_region_counts(sample)

Compare two samples

comparison <- compare_samples(resultA, resultB, threshold = 0.25)

Export results

write_experiment_results(result, "H3K4me3", "results.tsv")
write_comparison_results(comparison, "comparison.tsv")

---

📈 Outputs

Metric	Description
counts	Raw read count per region
lambda_g	Genome-wide background rate
lambda_t	Locally adjusted expected signal
p_value, p_adj	NB test p-values and FDR-adjusted values
score	log₂(Observed / Expected) enrichment
z_nb	NB-based z-score
DeltaEnrichment, DeltaZscore	Pairwise comparison metrics

---

💡 Contact

For questions, issues, or feature requests, please open a 👉 GitHub issue