semseeker

semseeker detects Stochastic Epigenetic Mutations (SEMs) — aberrant, sample-specific methylation changes that deviate from the expected population signal. Starting from a normalised beta-value (or M-value) matrix, semseeker identifies hypomethylated and hypermethylated mutations at single-probe resolution, aggregates them into lesions (genomic clusters of co-occurring mutations), and computes population-level delta metrics (ΔBeta, ΔIV, ΔEV, ΔRV) that summarise stochastic epigenetic variability per region.

Supported input technologies:

• Illumina microarrays: EPIC (850k), 450k, 27k — detected automatically
• Bisulfite sequencing (WGBS / RRBS) and long-read sequencing (ONT / PacBio): planned (see documents/requirements.md)

---

Installation

Install from GitHub using devtools:

install.packages("devtools")
library(devtools)
install_github("drake69/semseeker")

Dependencies

semseeker relies on several CRAN packages. Install them before the first use:

install.packages(c(
  "doFuture", "FactoMineR", "factoextra", "FSA", "fst",
  "future", "future.apply", "ggplot2", "gridExtra", "gtools",
  "Hmisc", "lqmm", "progressr", "reshape2", "zoo",
  "VennDiagram", "effsize", "pwr", "combinat", "philentropy",
  "parallelly", "knitr", "lmtest", "usethis", "wordcloud",
  "WebGestaltR", "sqldf", "caret", "openai", "glmnet"
))

---

Quick Example

Step 1 — Load and normalise IDAT files with ChAMP

library(ChAMP)

idat_folder  <- "~/source_idat/"
result_folder <- "~/result/"

myLoad <- champ.load(
  directory      = idat_folder,
  method         = "minfi",
  methValue      = "B",
  autoimpute     = TRUE,
  filterDetP     = TRUE,
  ProbeCutoff    = 0,
  SampleCutoff   = 0.1,
  detPcut        = 0.01,
  filterBeads    = TRUE,
  beadCutoff     = 0.05,
  filterNoCG     = TRUE,
  filterSNPs     = TRUE,
  filterMultiHit = TRUE,
  filterXY       = TRUE,
  arraytype      = "EPIC"   # or "450K"
)

myNorm <- champ.norm(
  beta       = myLoad$beta,
  rgSet      = myLoad$rgSet,
  mset       = myLoad$mset,
  resultsDir = result_folder,
  method     = "SWAN",
  arraytype  = "EPIC",
  cores      = parallel::detectCores() - 1
)

saveRDS(myNorm, "~/normalizedData.rds")

Step 2 — Run semseeker

library(semseeker)

normalizedData <- readRDS("~/normalizedData.rds")
sample_sheet   <- read.csv2("~/sample_sheet.csv")

semseeker(
  sample_sheet  = sample_sheet,
  signal_data   = normalizedData,
  result_folder = "~/semseeker_result/"
)

Results (BED files, pivot tables, delta-metric summaries) are written to result_folder.

---

Complete Example

A fully worked example using public data from Gene Expression Omnibus (GEO) is available in the repository's example/ folder.

---

Input Requirements

Beta-value matrix

• Rows = probes (Illumina probe IDs, e.g. cg...)
• Columns = samples
• Values = beta values in [0, 1] or M-values (detected automatically)

Sample sheet

A data frame (or CSV readable with read.csv2) with at minimum these columns:

Column	Description
Sample_ID	Unique sample identifier (must match matrix column names)
Sample_Name	Human-readable label
Sample_Group	Must be one of "Case", "Control", "Reference"

Note: If no Reference population is available, duplicate the Control rows and assign Sample_Group = "Reference" to the duplicates.

Note: Sample_ID values are uppercased internally by semseeker (name_cleaning()); output file names and pivot column headers reflect the uppercased form.

---

Output

For each sample semseeker writes to result_folder/Data/:

• *_MUTATIONS_HYPO.bed / *_MUTATIONS_HYPER.bed — probe-level mutations
• *_LESIONS_HYPO.bed / *_LESIONS_HYPER.bed — genomic lesion clusters
• *_DELTA*.bed — ΔBeta, ΔIV, ΔEV, ΔRV per region
• Pivot tables aggregating all samples per marker

---

Citation

If you use semseeker in published work, please cite the package:

citation("semseeker")

A Zenodo DOI will be registered for each release tag (see documents/requirements.md).