consistent use of spaces (rather than mixing tabs and spaces, e.g., in rjmcmcNucleo.cpp) and formatting would help readability.

Please either populate the NEWS file in a format that can be parsed
by news() (see, e.g., the NEWS file in GenomicRanges or remove
this file.

Add missing unit tests.

Add comparison metrics.

postMerge(): avoid 1:nbOverlap in for loops, which has
surprising results when nbOverlap == 0; use seq_len(nbOverlap)
instead. It seems like (parts of?) this iteration could be avoided
by vectorization?

It's sufficient to include Authors@R, without the Author: or
Maintainer: fields; these are generated when the package is built
(R CMD build RJMCMCNucleosomes).

minor typos in documentation, e.g. RJMCMCNucleosomes.R:48 "froward"

Add a section for rjmcmcCHR() function in vignette

1- segmentation(). The 'dataIP' argument is described 'reads that
need to be segmented'. Reads are represented most readily as
GAlignments() or GAlignmentPairs(), from the
GenomicAlignments package; should these be supported as input?

2- segmentation(). The example might more easiy coerce to GRanges
with
as(syntheticNucleosomeReads$dataIP, "GRanges")

3- segmentation() returns a simple list, but would be more convenient
as a GRangesList(), the equivalent of
GRangesList(segmentation(...))

4- segmentation() seems to assume that
length(unique(seqnames(dataIP))) == 1L?

5- segmentation() it seems like you're splitting based on overlap, so
a much more efficient implementation is
starts = seq(posMin, posMax, by = (maxLength - (zeta + delta)))
subject = GRanges(seqlevels(dataIP), IRanges(starts, width=maxLength)
hits = findOverlaps(dataIP, subject)
splitAsList(dataIP[queryHits(hits)], subjectHits(hits))

mergeAllRDSFiles() the pattern used here is 'copy-and-append',
which scales quadratically with the size of data (all existing data
is copied each time a new data element is added). If the data is
large, consider a better strategy, e.g., 'pre-allocate-and-fill',
e.g.,

mu <- setNames(vector("list", length(arrayOfFiles)), arrayOfFiles)
for (fileName in arrayOfFiles)
...
mu[[fileName]] <- data$mu
...
mu <- unlist(mu, use.names=FALSE

1 - It seems like the major functions should inter-operate with the
GenomicRanges package and in particular GRanges() class. The first
two arguments of rjmcm() should be (minimally, support) GRanges
instances. The return value should be a GRanges, perhaps with
additional information in the metadata().

2 - This leads to use of S4 classes that extend GRanges, rather than ad
hoc S3 classes like rjmcmNucleosomes. The classes do not have to be
complicated, e.g.,
setClass("RJMCMNucleosomes", contains="GRanges")
with minimal additional methods, e.g., perhaps show().

Add paired-end treatment.

rjmcmCHR() I suggest you use BiocParallel::bplappy() rather
than mclapply(), so that Windows users also get parallel
evaluation.

a <- bplapply(1:nbSeg, FUN = runCHR, seg, niter = nbrIterations,
kmax = kMax, lambda = lambda, ecartmin = minInterval,
ecartmax = maxInterval, minReads = minReads,
adaptNbrIterations = adaptIterationsToReads,
vSeed = vSeed, saveAsRDS = saveAsRDS)