macs (version 1.0)

tarv: Recursive partitioning based classification trees for risk profile and diagnosis

Description

Fit a ctree model

Usage

tarv(pheno, geno, formula, method = c("entropy", "gini"),
                  family = c("binomial", "gaussian"),
                  direction = c("both", "positive", "negative"),
                  alpha = 0.01, cost = NULL)

Arguments

pheno

defines the phenotypic data, serving as the response variable.

geno

defines the genetic variants.

formula

specifies the model structure such as 'disease~sex+race'.

method

is the same as that in rtree.

family

specifies the regression method: 'binomial' for the logistic regression for a binary phenotype and 'gaussian' for the linear regression with a normally distributed phenotype.

direction

offers the option for the inclusion of genetic markers. When tarv processes the original genetic markers, it may rank their univariate effects according to their t-values in three ways, namely t (positive), -t (negative), and |t| (both). We recommend both to avoid any potential confusion which means that more statistically significant genetic markers would be considered favorably regardless of their directions of the effect on the phenotype.

alpha

is the same as that in rtree.

cost

is the same as that in rtree.

Value

nnd

the total number of nodes in the tree.

dt

the sequence number of a left daughter node for each internal node.

pt

the sequence number of the parent node for any daughter node.

spv

the splitting variable used to split a given node.

spvl

the cut-off value of the splitting variable above.

final_counts

the table that contains the number of observations in each node.

varcatg

a numerical indicator for the category of each variable. Value '-1' points to the response variable, '1' to oridinal variables, an integer greater than 1 to a nominal variable with the number of levels equal to the integer.

nodeclass

the class membership of a terminal node which depends on the choice of the misclassification cost.

p_value

the p-value of the chi-square test performed at each internal node. It forms the basis to prune the offspring nodes of any internal node. More details in Recursive Partitioning and Applications [Zhang and Singer].

call

the call by which this object is generated.

learning.data

the data that are actually used in rtree and tarv.

Details

'pheno' and 'geno' are processed via TRAV[Song and Zhang 2014] to create new covariates and then these new virables are used to create a tree construction. So it has a similar input and output form with rtree. The object derives from rtree and ctree function are both 'ctree' object, and can be used to plot.ctree and predict.ctree.

References

Zhang, H. and Singer, B. (1999), Recursive partitioning in the health sciences, Springer Verlag.

Song C. and Zhang H.(2014), Tree-based Analysis of Rare Variants Identifying Risk Modifying Variants in CTNNA2 and CNTNAP2 for Alcohol Addiction.

See Also

plot.ctree, forecast.ctree, rtree

Examples

# NOT RUN {
library(macs)
set.seed(123)
sex <- rbinom(1000, 1, 0.5)
race <- sample(1:3, 1000, replace = TRUE, prob = c(0.4, 0.4, 0.2))
gg <- replicate(100, rbinom(1000, 1, runif(1, 0.005, 0.05)))
annotation <- paste("gene", rep(1:5, each = 20), sep = "")
causal <- rbinom(40, 1, 0.8)
x1 <- rowSums(gg[, 1:20][, causal[1:20] > 0]) > 0
x2 <- rowSums(gg[, 21:40][, causal[21:40] > 0]) > 0
xb <- sex * 0.2 + (race == 2) * 0.2 + x1 * 0.6 + x2 * 0.8 - 0.7
r <- rbinom(1000, 1, exp(xb)/(1+exp(xb)))
pheno <- data.frame(disease = r, sex = sex, race = as.factor(race))
geno <- t(gg)
rownames(geno) <- annotation

result <- tarv(pheno, geno, formula = "disease~sex+race",
                 method = "entropy", family = "binomial",
                 direction = "both", alpha = 0.01, cost = c(1, 1))
plot(result)
# }