plrs: Fit a (constrained) piecewise linear regression spline

Description

The function fits a piecewise linear regression spline to explain gene expression by the segmented DNA copy number. The called copy number values are used as a template for model building.

Usage

plrs(expr, cghseg, cghcall=NULL, probloss = NULL, probnorm = NULL,
probgain = NULL, probamp = NULL, knots = NULL, continuous = FALSE,
constr = TRUE, constr.slopes = 2, constr.intercepts = TRUE,
min.obs = 3, discard.obs = TRUE)

Arguments

expr

Vector of gene expression values

cghseg

Vector of segmented copy number values

cghcall

Vector of called copy number values. If not provided, we are reduced to a simple linear model.

probloss

Vector of call probabilities associated with state "loss". Default is NULL.

probnorm

Vector of call probabilities associated with state "normal". Default is NULL.

probgain

Vector of call probabilities associated with state "gain". Default is NULL.

probamp

Vector of call probabilities associated with state "amplification". Default is NULL.

knots

knots or change points. If NULL (default), there are estimated. See details.

continuous

Logical, whether the model is continuous (no jump) or not.

constr

Logical, whether the model is constrained or not. (this has been implemented to turn on and off easily the constraints)

constr.slopes

Type of non-negativity constraints applied on slopes. Either 1 or 2 (default). See details.

constr.intercepts

If TRUE (default) jumps from state to state are also constrained to be non-negative

min.obs

See modify.conf

discard.obs

See modify.conf

Value

plrs-class

Details

If cghcall=NULL, discrete copy number values are omitted, which results in fitting a simple linear model.

If constr.slopes=1, all slopes are constrained to be non-negative. If constr.slopes=2, the slope associated with state "normal" is constrained to be non-negative and all others are forced to be at least equal to the latter.

Two methods are implemented for the estimation of knots. If call probabilities are provided, a knot is determined so that the sum of (the two adjacent) states membership probabilities is maximized. Otherwise, this is defined as the midpoint of the interval between the two consecutive states.

The constrained least squares problem is solved using function solve.QP of package quadprog.

Examples

Run this code


# Simulate data
sim <- plrs.sim(n=80, states=4, sigma=0.5)

# Fit a model 
model <- plrs(expr=sim$expr, cghseg=sim$seg, cghcall=sim$cal)
model

# Methods
coef(model)
effects(model)
fitted(model)
knots(model)
model.matrix(model)
plot(model)
predict(model, newcghseg=seq(0,5, length.out=100))   
residuals(model)
summary(model)