callPeaksUnivariateAllChr: Fit a Hidden Markov Model to a ChIP-seq sample.
In chromstaR: Combinatorial and Differential Chromatin State Analysis for ChIP-Seq Data

Description Usage Arguments Details Value Author(s) See Also

Fit a HMM to a ChIP-seq sample to determine the modification state of genomic regions, e.g. call peaks in the sample.

callPeaksUnivariateAllChr(
  binned.data,
  control.data = NULL,
  eps = 0.01,
  init = "standard",
  max.time = NULL,
  max.iter = NULL,
  num.trials = 1,
  eps.try = NULL,
  num.threads = 1,
  read.cutoff = TRUE,
  read.cutoff.quantile = 1,
  read.cutoff.absolute = 500,
  max.mean = Inf,
  post.cutoff = 0.5,
  control = FALSE,
  keep.posteriors = FALSE,
  keep.densities = FALSE,
  verbosity = 1
)

`binned.data`	A `GRanges-class` object with binned read counts or a file that contains such an object.
`control.data`	Input control for the experiment. A `GRanges-class` object with binned read counts or a file that contains such an object.
`eps`	Convergence threshold for the Baum-Welch algorithm.
`init`	One of the following initialization procedures: `standard` The negative binomial of state 'unmodified' will be initialized with `mean=mean(counts)`, `var=var(counts)` and the negative binomial of state 'modified' with `mean=mean(counts)+1`, `var=var(counts)`. This procedure usually gives the fastest convergence. `random` Mean and variance of the negative binomials will be initialized with random values (in certain boundaries, see source code). Try this if the `'standard'` procedure fails to produce a good fit. `empiric` Yet another way to initialize the Baum-Welch. Try this if the other two methods fail to produce a good fit.
`max.time`	The maximum running time in seconds for the Baum-Welch algorithm. If this time is reached, the Baum-Welch will terminate after the current iteration finishes. The default `NULL` is no limit.
`max.iter`	The maximum number of iterations for the Baum-Welch algorithm. The default `NULL` is no limit.
`num.trials`	The number of trials to run the HMM. Each time, the HMM is seeded with different random initial values. The HMM with the best likelihood is given as output.
`eps.try`	If code num.trials is set to greater than 1, `eps.try` is used for the trial runs. If unset, `eps` is used.
`num.threads`	Number of threads to use. Setting this to >1 may give increased performance.
`read.cutoff`	The default (`TRUE`) enables filtering of high read counts. Set `read.cutoff=FALSE` to disable this filtering.
`read.cutoff.quantile`	A quantile between 0 and 1. Should be near 1. Read counts above this quantile will be set to the read count specified by this quantile. Filtering very high read counts increases the performance of the Baum-Welch fitting procedure. However, if your data contains very few peaks they might be filtered out. If option `read.cutoff.absolute` is also specified, the minimum of the resulting cutoff values will be used. Set `read.cutoff=FALSE` to disable this filtering.
`read.cutoff.absolute`	Read counts above this value will be set to the read count specified by this value. Filtering very high read counts increases the performance of the Baum-Welch fitting procedure. However, if your data contains very few peaks they might be filtered out. If option `read.cutoff.quantile` is also specified, the minimum of the resulting cutoff values will be used. Set `read.cutoff=FALSE` to disable this filtering.
`max.mean`	If `mean(counts)>max.mean`, bins with low read counts will be set to 0. This is a workaround to obtain good fits in the case of large bin sizes.
`post.cutoff`	False discovery rate. codeNULL means that the state with maximum posterior probability will be chosen, irrespective of its absolute probability (default=codeNULL).
`control`	If set to `TRUE`, the binned data will be treated as control experiment. That means only state 'zero-inflation' and 'unmodified' will be used in the HMM.
`keep.posteriors`	If set to `TRUE` (default=`FALSE`), posteriors will be available in the output. This is useful to change the post.cutoff later, but increases the necessary disk space to store the result.
`keep.densities`	If set to `TRUE` (default=`FALSE`), densities will be available in the output. This should only be needed debugging.
`verbosity`	Verbosity level for the fitting procedure. 0 - No output, 1 - Iterations are printed.

The Hidden Markov Model which is used to classify the bins uses 3 states: state 'zero-inflation' with a delta function as emission densitiy (only zero read counts), 'unmodified' and 'modified' with Negative Binomials as emission densities. A Baum-Welch algorithm is employed to estimate the parameters of the distributions. Please refer to our manuscript at http://dx.doi.org/10.1101/038612 for a detailed description of the method.

A uniHMM object.

Aaron Taudt, Maria Coome Tatche

uniHMM, callPeaksMultivariate

chromstaR documentation built on Nov. 8, 2020, 8:29 p.m.