02-cnvPosterior: Class "CNVPosterior"

In DeepCNV: Exploiting Normal Contamination to Infer Copy Number from Deep Sequencing

Description

Represents the posterior distribution used to infer copy number from targeted deep sequencing.

Usage

makeCNVPosterior(obs, prior)
## S4 method for signature 'CNVPosterior,missing'
plot(x, place="topright", lwd=1, ...)
## S4 method for signature 'CNVPosterior'
hist(x, place="topright", lwd=1, ...)
## S4 method for signature 'CNVPosterior'
summary(object, ...)
## S4 method for signature 'CNVPosteriorSummary'
as.data.frame(x, row.names = NULL, optional = FALSE, ...)

Arguments

obs	Observed read-count data
prior	Prior distribution, represented using the class CNVPrior.
object	object of class CNVPosterior
x	object of class CNVPosterior
place	Character string; where to place the figure legend
lwd	Line width parameter.
row.names	See as.data.frame
optional	See as.data.frame
...	extra arguments for generic or plotting routines

Details

The DeepCNV class is used to fit a Bayesian model to targeted sequencing data from one or a few genes in order to draw inferences about possible copy number changes. Basically, we assume that the observed data consists of a list of triples (K, N, V), one for each variant in a gene. Here K is the number of variant reads, N is the total number of reads, and V is the type of each variant (either a known SNP or a somatic mutation). We model (K, N) using a binomial distribution, where the 'success' parameter depends (in a deterministic way) on the unknown parameters of interest: the fraction ν of normal cells in the sample and the copy number state (Normal, Deleted, or Gained).

The prior distribution consists of a continuous (by default, Beta(α, β)) distribution on ν and a discrete distribution on the copy number state S, which are stored ina n object of the CNVPrior class. The CNVariant computes the success parameter φ as a function of the observed data (K,N,V). Then φ and thbe observed data are passed on to the cnvLikelihood function, whcih computes the binomial log-likelihood.

We compute the posterior distribution essentially by brute force, in the function makeCNVPosterior. We choopse regularly spaced grid of points on the inerval (0,1) and evaluate the log-likelihood at every grid point foir each conceivable copy number state. These likelihoods are combined with the prior distribution by teh usual application of Bayes Rule.

One complication arises from the variant type V, which can be observed with error. Additioal complications arise because the variant (whether SNP or mutation) can turn out to be either the major or minor allele. These difficulties are mostly hidden from the user, and resolved by rfeplacing an overabundaqnce of states with the maximu a posteriori selection at each variant. This design decision afects the structure of the class, which compute multiple forms of the likelihood:

1. hiddenloglike, for all hidden states

2. snploglike, for each variant separately, collapsiong hidden states

3. loglike, for the unified gene

and posterior distribution:

1. snppost, for eqach variant separately

2. posterior, for the unified gene

Value

The makeCNVPosterior constructor returns a valid object of the class.

Slots

hiddenloglike:

Log-likelihood using the hidden expanded discrete model

snploglike:

Log-likelihood for each SNP analyzed one at a time

snppost:

Posterior distribution of parameters for each SNP analyzed one at a time

loglike:

Log-likelihood merging data from all SNPs

posterior:

Posterrio distribution merging data from all SNPs

calls:

List of maximu a posterior variant types used to collapse the hidden model

observed:

Observed read data

prior:

Prior distribution

Methods

plot(x, ...)

Prints all the information in the object

hist(x, ...)

Prints all the information in the object

summary(object, ...)

Writes out a summary of the object

Author(s)

Kevin R. Coombes [e-mail]

See Also

CNVPosterior

Examples

prior <- setCNVPrior(alpha=1.2, beta=4.8, pAbnormal=0.6)
dataset <- simReads(2, 7, 0.17, "Normal")
posterior <- makeCNVPosterior(dataset, prior)
s <- summary(posterior)
s
as.data.frame(s)
plot(posterior)
hist(posterior)

DeepCNV documentation built on May 2, 2019, 5:23 p.m.