casper (version 2.6.0)

modelPrior: Set prior distribution on expressed splicing variants.

Description

Set prior on expressed splicing variants using the genome annotation contained in a knownGenome object.

The prior probability of variants V1,...,Vn being expressed depends on n, on the number of exons in each variant V1,...,Vn and the number of exons in the gene. See the details section.

Usage

modelPrior(genomeDB, maxExons=40, smooth=TRUE, verbose=TRUE)

Arguments

genomeDB

Object of class knownGenome

maxExons

The prior distribution is estimated for genes with 1 up to maxExons exons. As there are fewer genes with many exons, the prior parameters are estimated poorly. To avoid this common estimate is used for all genes with more than maxExons exons

smooth

If set to TRUE the estimated prior distribution parameters for the number of exons in a gene are smoothed using Generalized Additive Models. This step typically improves the precision of the estimates, and is only applied to genes with 10 or more exons.

verbose

Set to TRUE to print progress information.

Value

List with 2 components.

nvarPrior

List with prior distribution on the number of expressed variants for genes with 1,2,3... exons. Each element contains the truncated Negative Binomial parameters, observed and predicted frequencies (counting the number of genes with a given number of variants).

nexonPrior

List with prior distribution on the number of exons in a variant for genes with 1,2,3... exons. Each element contains the Beta-Binomial parameters, observed and predicted frequencies (counting the number of variants with a given number of exons)

Details

The goal is to set a prior that takes into account the number of annotated variants for genes with E exons, as well as the number of exons in each variant. Suppose we have a gene with E exons. Let V_1,...,V_n be n variants of interest and let |V_1|,...,|V_n| be the corresponding number of exons in each variant. The prior probability of variants V_1,...,V_n being expressed is modeled as

P(V_1,...,V_n|E)= P(n|E) P(|V_1| |E) ... P(|V_n| |E)

where P(n|E)= NegBinom(n; k_E, r_E) I(0 < n < 2^E) and P(|V_i| |E)= BetaBinomial(|V_i|-1; E-1, alpha_E, beta_E). The parameters k_E, r_E, alpha_E, beta_E depend on E (the number of exons in the gene) and are estimated from the available annotation via maximum likelihood. Parameters are estimated jointly for all genes with E>= maxExons in order to improve the precision. For smooth==TRUE, alpha_E and beta_E are modeled as a smooth function of E by calling gam and setting the smoothing parameter via cross-validation. Estimates for genes with E>=10 are substituted by their smooth versions, which typically helps improve stability in the estimates.

Examples

data(hg19DB)
mprior <- modelPrior(hg19DB, maxExons=10)

##Prior on number of expressed variants
##Genes with 2 exons
##mprior$nvarPrior[['2']]
##Genes with 3 exons
##mprior$nvarPrior[['3']]

##Prior on the number of exons in an expressed variant
##Genes with 2 exons
##mprior$nexonPrior[['2']]
##Genes with 3 exons
##mprior$nexonPrior[['3']]