net.infer.ST: Inferring functionally related proteins with self training

In dongminjung/PPInfer: Inferring functionally related proteins using protein interaction networks

Description

This function is the self-training version of net.infer. The function net.infer is the special case of net.infer.ST where a single iteration is conducted.

Usage

net.infer.ST(target, kernel, top = NULL, C = 1, nu = 0.2,
            epsilon = 0.1, cache1 = 40, tol1 = 0.001, shrinking1 = TRUE,
            cache2 = 40, tol2 = 0.001, shrinking2 = TRUE, thrConf = 0.9,
            maxIts = 10, percFull = 1, verbose = FALSE) 

Arguments

target	set of interesting proteins or target class
kernel	the regularized Laplacian matrix for a graph
top	number of top proteins most closely related to target class (default: all proteins except for target and pseudo-absence class)
C	cost of constraints violation for SVM (default: 1)
nu	The nu parameter for OCSVM (default: 0.2)
epsilon	epsilon in the insensitive-loss function for OCSVM (default: 0.1)
cache1	cache memory in MB for OCSVM (default: 40)
tol1	tolerance of termination criterion for OCSVM (default: 0.001)
shrinking1	option whether to use the shrinking-heuristics for OCSVM (default: TRUE)
cache2	cache memory in MB for SVM (default: 40)
tol2	tolerance of termination criterion for SVM (default: 0.001)
shrinking2	option whether to use the shrinking-heuristics for SVM (default: TRUE)
thrConf	A number between 0 and 1, indicating the required classification confidence for an unlabelled case to be added to the labelled data set with the label predicted predicted by the classification algorithm (default: 0.9)
maxIts	The maximum number of iterations of the self-training process (default: 10)
percFull	A number between 0 and 1. If the percentage of labelled cases reaches this value the self-training process is stoped (default: 1)
verbose	A boolean indicating the verbosity level of the function. (default: FALSE)

Value

list	list of a target class used in the model
error	training error
top	top proteins
score	decision values for top proteins

Author(s)

Dongmin Jung, Xijin Ge

See Also

self.train

Examples

data(litG)
litG <- igraph.from.graphNEL(litG)
sg <- decompose(litG, min.vertices = 50)
sg <- sg[[1]]
K <- net.kernel(sg)
litG.infer.ST <- net.infer.ST(names(V(sg))[1:10], K, top=20)

dongminjung/PPInfer documentation built on May 15, 2019, 10:41 a.m.