eagle: Run the full EAGLE procedure.

In davidaknowles/eagle: Environment-ASE through Generalized LinEar modeling

Description

The usual usage is to 1) fit the null model across exonic SNPs, while learning the overdispersion hyperparameters, 2) fit the alternative models holding these hyperparameters fixed 3) perform likelihood ratio tests for each locus

Usage

eagle(alt,n,xFull,xNull,max.its=1000,tol=10.0,debug=F,learn.rev=T,rev=1.0,...
	traceEvery=1,rev.model="global",null.first=T,coeff.reg=0.0,...
	return.aux.variables=F,storeAllCoeffs=F,repRep=1)

Arguments

alt	list (over exonic SNPs) of alternative read counts
n	list (over exonics SNPs) of total read counts
xFull	list of design matrices for the alternative hypothesis (e.g. including environment)
xNull	list of design matrices for the null hypothesis
max.its	maximum iterations of EM to run
tol	consider EM to have converged if the change in the lower bound is less than this
debug	output debugging information? Also performs additional checks.
learn.rev	whether to learn the overdispersion hyperparameters a,b
rev	initial random effect variance
traceEvery	traceEvery: how often to output convergence info to stdout
rev.model	rev.model: one of c("global","regression","local","local.regression"). global: single fixed random effect variance across all exonic SNPs (not recommended). regression: random effect variance is log-linear in the log average read depth (not recommended). local: random effect variance at exonic SNP s is v_s ~ InverseGamma(a,b) [recommended]. local.regression: log(rev) ~ N(mt+c,v) [gives similar results to "local" but not as well calibrated]
nullFirst	whether to run the null or alternative model first. Only matters if learning the overdispersion hyperparamters, in which case these are learnt on whichever is run first, and fixed for the second.
coeff.reg	whether to regularize the regression coefficents with a term -coef.reg |beta|^2
return.aux.variables	whether to return the auxiliary variables g (only used for debugging)
storeAllCoeffs	whether to store coefficients throughout the EM algorithm (for debugging)
repRep	1/v in the local.regression overdispersion model log(rev) ~ N(mt+c,v)
rerunFirst	whether to rerun the first model after learning hyperparameters (not used)
learnRepRep	whether to learn v in the local.regression model
learnBetas	whether to learn the regression coefficients

Value

The returned list is the same as for eagle.helper:

p.values	$p$-values for the likelihood ratio tests.
q.values	Corresponding $q$-values calculated using Benjamini-Hochberg FDR.
res.full	Results for the full (alternative) design matrices. See eagle.vem for details.
res.null	Results for the null design matrices.
settings	Settings used, see eagle.settings
timeFirst	Computation time for first run.
timeSecond	Computation time for second run.
res.firstOld	If the first run was re-run after fitting hyperparameters, then these are the results of that initial run.

Author(s)

David A. Knowles

davidaknowles/eagle documentation built on May 14, 2019, 9:35 p.m.