Fit the exomeCopy or exomeCopyVar model to the observed counts.

Share:

Description

Fits a hidden Markov model to observed read counts using positional covariates. It returns an object containing the fitted parameters and the Viterbi path, the most likely path of hidden states, which is the predicted copy count at each window.

exomeCopy is designed to run on read counts from consecutive genomic ranges on a single chromosome, as it tries to identify higher or lower read depth relative to a baseline. Please see the vignette for an example of how to prepare input data for exomeCopy, how to loop the function over multiple chromosomes and samples, and how to extract the resulting predicted CNVs.

exomeCopy requires as input a RangedData object containing read counts in genomic ranges along with the covariates. Some convenience functions are provided for preparing input for exomeCopy:

  1. subdivideGRanges, to subdivide a GRanges object containing the genomic ranges of the targeted region into genomic ranges of nearly equal width,

  2. countBamInGRanges, to count the number of read starts from a BAM read mapping file in a GRanges object,

  3. getGCcontent, to get the GC-content in the ranges given a FASTA file of the reference sequence,

  4. generateBackground, to calculate median normalized read depth over a set of control samples, and also any statistic over normalized read depth.

Usage

1
2
3
  exomeCopy(rdata, sample.name, X.names, Y.names, fit.var=FALSE, reltol
  = 0.0001, S = 0:4, d = 2, goto.cnv = 1e-4, goto.normal = 1/20,
init.phi="norm")    

Arguments

rdata

A RangedData object with the sample counts and positional covariates over the genomic ranges.

sample.name

The name of the value column of rdata with the sample read counts.

X.names

The names of the value columns of rdata with covariates for estimating mu.

Y.names

(optional) the names of the value columns of rdata with covariates for estimating phi, only required if fit.var = TRUE.

fit.var

A logical, whether the model should fit the overdispersion parameter phi with a linear combination of covariates (exomeCopyVar) or with a scalar (exomeCopy). Defaults to FALSE (exomeCopy).

reltol

The relative tolerance for convergence used in the optim function for optimizing the parameter settings. From testing, the default value was sufficient for fitting parameters, but lower relative tolerances can be used.

S

A vector of possible copy numbers for the different states.

d

The expected copy number for the normal state. This should be set to 2 for autosomes and 1 for haploid data.

goto.cnv

The initial setting for probability to transfer to a CNV state.

goto.normal

The initial setting for probability to transfer to a normal state.

init.phi

Either "norm" or "counts": initialize phi with the moment estimate using residuals from a linear model of read counts on covariates or with the raw counts.

Details

exomeCopy fits transitional and emission parameters of an HMM to best explain the observed counts of a sample from exome or targeted sequencing. The set of underlying copy number states, S, in the sample must be provided before running the algorithm.

The emission probabilities are given as a negative binomial distribution using positional covariates, such as log background read depth, quadratic terms for GC-content, and range width, which are stored in a matrix X. Optionally, for fitting the variance of the distribution, the standard deviation and/or variance of the background set can be included in a matrix Y. All covariates are normalized within exomeCopy for improved optimization.

For the observed count at range t, O_t, the emission probability is given by:

O_t ~ NB(mu_ti, phi)

The mean parameter mu_ti is given by:

mu_ti = (S_i / d) e^(x_t* beta)

Here S_i is the i-th possible copy number state, d is the expected background copy number (d = 2 for diploid sequence), and beta is a vector of coefficients fitted by the model. x_t* is the t-th row of the matrix X.

mu must be positive, so it is replaced with a small positive number if the value is less than zero.

For exomeCopyVar, which also fits the variance, the emission probability includes a location-dependent dispersion parameter.

where

log(phi_t) = y_t* gamma

or a small positive number if this is less than zero.

Two transition probabilities are fitted in the model: the probabilities of transitioning to a normal state and to a CNV state.

exomeCopy calls negLogLike to evaluate the likelihood of the HMM. The parameters are fit using Nelder-Mead optimization with the optim function on the negative likelihood. The Viterbi path is calculated by calling viterbiPath.

Value

Returns an ExomeCopy-class object. See this page for the slot descriptions. Also see the vignette and copyCountSegments on how to extract segments.

Author(s)

Michael Love

References

Love, Michael I.; Mysickova, Alena; Sun, Ruping; Kalscheuer, Vera; Vingron, Martin; and Haas, Stefan A. (2011) "Modeling Read Counts for CNV Detection in Exome Sequencing Data," Statistical Applications in Genetics and Molecular Biology: Vol. 10 : Iss. 1, Article 52. DOI: 10.2202/1544-6115.1732 http://cmb.molgen.mpg.de/publications/Love_2011_exomeCopy.pdf.

References for HMM algorithms and use of HMM for segmentation of genomic data by copy number:

Rabiner, L. R. (1989): "A tutorial on hidden Markov models and selected applications in speech recognition," Proceedings of the IEEE, 77, 257, 286, http://dx.doi.org/10.1109/5.18626.

Fridlyand, J., A. M. Snijders, D. Pinkel, D. G. Albertson, and Jain (2004): "Hidden Markov models approach to the analysis of array CGH data," Journal of Multivariate Analysis, 90, 132, 153, http://dx.doi.org/10.1016/j.jmva.2004.02.008.

Marioni, J. C., N. P. Thorne, and S. Tavare (2006):"BioHMM: a heterogeneous hidden Markov model for segmenting array CGH data." Bioinformatics, 22, 1144, 1146, http://view.ncbi.nlm.nih.gov/pubmed/16533818.

See Also

ExomeCopy-class subdivideGRanges countBamInGRanges copyCountSegments plot.ExomeCopy negLogLike IRanges RangedData

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
## The following is an example of running exomeCopy on simulated
## read counts using the model parameters defined above.  For an example
## using real exome sequencing read counts (with simulated CNV) please
## see the vignette.

## create RangedData for storing genomic ranges and covariate data
## (background, background stdev, GC-content)

m <- 5000
rdata <- RangedData(IRanges(start=0:(m-1)*100+1,width=100),
space=rep("chr1",m), universe="hg19", log.bg=rnorm(m), log.bg.var=rnorm(m),
gc=runif(m,30,50))  

## create read depth distributional parameters mu and phi
rdata$gc.sq <- rdata$gc^2
X <- cbind(bg=rdata$log.bg,gc=rdata$gc,gc.sq=rdata$gc.sq)
Y <- cbind(bg.sd=rdata$log.bg.var)
beta <- c(5,1,.01,-.01)
gamma <- c(-3,.1)
rdata$mu <- exp(beta[1] + scale(X) %*% beta[2:4])
rdata$phi <- exp(gamma[1] + scale(Y) %*% gamma[2])

## create observed counts with simulated heterozygous duplication
cnv.nranges <- 200
bounds <- (round(m/2)+1):(round(m/2)+cnv.nranges)
O <- rnbinom(nrow(rdata),mu=rdata$mu,size=1/rdata$phi)
O[bounds] <- O[bounds] + rbinom(cnv.nranges,prob=0.5,size=O[bounds])
rdata[["sample1"]] <- O

## run exomeCopy() and list segments
fit <- exomeCopy(rdata,"sample1",X.names=c("log.bg","gc","gc.sq"))

# an example call with variance fitting.
# see paper: this does not necessarily improve the fit
fit <- exomeCopy(rdata,"sample1",X.names=c("log.bg","gc","gc.sq"),
                 Y.names="log.bg",fit.var=TRUE)

## see man page for copyCountSegments() for summary of
## the predicted segments of constant copy count, and
## for plot.ExomeCopy() for plotting fitted objects