gsynth.sample: Sample a synthetic genome from a trained Markov model

In misha: Toolkit for Analysis of Genomic Data

Sample a synthetic genome from a trained Markov model

Description

Generates a synthetic genome by sampling from a trained stratified Markov model. The generated genome preserves the k-mer statistics of the original genome within each stratification bin.

Usage

gsynth.sample(
  model,
  output_path = NULL,
  output_format = c("misha", "fasta", "vector"),
  mask_copy = NULL,
  preserve_n = TRUE,
  seed = NULL,
  intervals = NULL,
  n_samples = 1,
  bin_merge = NULL,
  cell_merge = NULL
)

Arguments

model	A gsynth.model object from gsynth.train
output_path	Path to the output file (ignored when output_format = "vector")
output_format	Output format: "misha": .seq binary format (default) "fasta": FASTA text format "vector": Return sequences as a character vector (does not write to file)
mask_copy	Optional intervals to copy from the original genome instead of sampling. Use this to preserve specific regions (e.g., repeats, regulatory elements) exactly as they appear in the reference. Regions not in mask_copy will be sampled using the Markov model. Note: mask_copy intervals should be non-overlapping and sorted by start position within each chromosome. Overlapping intervals may result in only the first overlapping region being copied, with subsequent overlaps skipped due to cursor advancement during sequential processing.
preserve_n	Logical; default TRUE. When TRUE, positions whose original reference is N (or n) are written to the output verbatim instead of being filled in with a sampled ACGT base. Case is preserved (so soft-masked n round-trips). mask_copy regions take precedence: inside a mask_copy interval the original byte is copied regardless. Set to FALSE to restore the pre-5.6.22 behavior of treating every position as something to sample.
seed	Random seed for reproducibility. If NULL, uses current random state.
intervals	Genomic intervals to sample. If NULL, uses all chromosomes.
n_samples	Number of samples to generate per interval. Default is 1. When n_samples > 1 and output_format = "fasta", headers include "_sampleN". When output_format = "vector", returns n_samples * n_intervals sequences.
bin_merge	Optional list of bin merge specifications to apply during sampling, one per dimension (length must equal model$n_dims). Each element should be: A list of merge specifications (same format as in gsynth.train: each spec is list(from = ..., to = ...)) Or NULL to use the bin mapping from training for that dimension This allows merging sparse bins at sampling time without re-training. Example for a 2D model: bin_merge = list( # Dimension 1: merge bins with values >= 0.8 to bin [0.7, 0.8) list(list(from = c(0.8, Inf), to = c(0.7, 0.8))), # Dimension 2: use training-time bin_map (no override) NULL )
cell_merge	Optional per-joint-cell redirect table, applied after bin_merge and after any sparse-bin uniform fallback. Each entry redirects one specific training cell to another specific training cell whose Cartesian position cannot be expressed as a per-axis merge. Format: a list where each element is list(from = c(v1, v2, ...), to = c(v1, v2, ...)), with one representative value per dimension; or a data frame with columns from_1, from_2, ..., to_1, to_2, .... Internally resolved via gsynth.cell_merge; at sampling time each source cell's CDF is replaced with the target cell's CDF (a pointer-level copy inside cdf_list — no matrix duplication and no change to the C++ hot path).

Details

FASTA index (.fai): When output_format = "fasta", the function also writes a samtools-compatible .fai file alongside the FASTA (byte offsets are tracked during the write loop, so this is effectively free). The index is suitable for any downstream tool that expects a samtools-indexed reference.

N bases during sampling: By default (preserve_n = TRUE) positions whose original reference is N are copied verbatim into the output, so gaps and centromeres remain N rather than being fabricated as ACGT. Where the sampler still has to fill an N-adjacent k-mer context (the first k bp inside an interval, or any k-mer window containing a preserved N), it falls back to uniform random base selection until a valid context is re-established. Similarly, if a bin has no learned statistics (sparse bin with NA CDF), uniform sampling is used for that position. Pass preserve_n = FALSE to recover the pre-5.6.22 behavior of sampling every position regardless of the reference.

Sparse bins: If the model has sparse bins (from min_obs during training), a warning is issued when sampling regions that fall into these bins. Consider using bin_merge to redirect sparse bins to well-populated ones.

Value

When output_format is "misha" or "fasta", returns invisible NULL and writes the synthetic genome to output_path. When output_format is "vector", returns a character vector of sequences (length = n_intervals * n_samples).

See Also

gsynth.train, gsynth.save

Examples

gdb.init_examples()

# Create virtual tracks
gvtrack.create("g_frac", NULL, "kmer.frac", kmer = "G")
gvtrack.create("c_frac", NULL, "kmer.frac", kmer = "C")
gvtrack.create("cg_frac", NULL, "kmer.frac", kmer = "CG")
gvtrack.create("masked_frac", NULL, "masked.frac")

# Define repeat mask (regions to preserve from original)
repeats <- gscreen("masked_frac > 0.5",
    intervals = gintervals.all(),
    iterator = 100
)

# Train model (excluding repeats from training)
model <- gsynth.train(
    list(expr = "g_frac + c_frac", breaks = seq(0, 1, 0.025)),
    list(expr = "cg_frac", breaks = c(0, 0.01, 0.02, 0.03, 0.04, 0.2)),
    mask = repeats,
    iterator = 200,
    min_obs = 1000
)

# Sample with mask_copy to preserve repeats from original genome
temp_dir <- tempdir()
synthetic_genome_file <- file.path(temp_dir, "synthetic_genome.fa")
gsynth.sample(model, synthetic_genome_file,
    output_format = "fasta",
    mask_copy = repeats,
    seed = 60427,
    bin_merge = list(
        list(list(from = 0.7, to = c(0.675, 0.7))),
        list(list(from = 0.04, to = c(0.03, 0.04)))
    )
)

misha documentation built on Aug. 24, 2026, 5:14 p.m.