GenoNet: Train the GenoNet model with dichotomous labels and features,...

In Ionita-Laza-lab/GenoNet: A Semi-Supervised Approach for Predicting Cell-Type Specific Functional Consequences of Non-Coding Variation

Description

Before predicting the functional consequences of genetic variation, this function fits the GenoNet model and preduces fitted parameters that will be used in function GenoNet.predict().

Usage

GenoNet(Y,X,beta.unlabeled,nfolds)

Arguments

Y	Dichotomous outcome variables or equivalently labels.
X	Features used for model training. If pre-calculated parameters from FunLDA are being used, X must have 1016 columns corresponding to the 1016 epigenetic features in 127 tissues or cell types, with the same order. See the complete list of features (ordered) at http://www.funlda.com/genonet/help.
beta.unlabeled	Pre-calculated parameters for from unlabeled data. Currenly pre-calculated parameters for 127 tissues based on 2 millions genome-wide unlabeled variants and their FunLDA scores (Backenroth D. et al. 2018) are available for direct use.
nfolds	Number of partitions in cross-validation.

Value

LogitBeta.labeled	Trained parameters from labeled data. Parameters are coefficients in a logistic model for labeled data.
LinearBeta.unlabeled	Trained parameters (same as the pre-calculated parameters) from unlabeled data. Parameters are coefficients in a linear model for unlabeled data.
phi	Fitted weight to combined predictions from labeled data and unlabeled data. Final prediction will be (1-phi)*labeled prediction + phi*unlabeled prediction. Smaller phi indicates more use of label data.

Examples

## GenoNet trains the model that will be used for prediction.
# Input: Y, X (covariates/features), pre-calculated parameters from unlabeled data

## GenoNet.predict predicts labels given a set of new features.
# Input: X (test data covariates/features), GenoNet.fit (fitted GenoNet model)

library(GenoNet)

#Load example data
data(GenoNet.example)
beta.FunLDA<-GenoNet.example$beta.FunLDA # pre-calculated parameters from unlabeled data
training.data<-GenoNet.example$training.data # example training data
test.data<-GenoNet.example$test.data # example test data

beta.unlabeled<-beta.FunLDA[,'E118'] # tissue E118, HepG2 Hepatocellular Carcinoma Cells
Y.training<-training.data[,4];Y.test<-test.data[,4]
X.training<-training.data[,-c(1:4)];X.test<-test.data[,-c(1:4)]

#fit GenoNet
GenoNet.fit<-GenoNet(Y.training,X.training,beta.unlabeled,nfolds=10)
Y.predict<-GenoNet.predict(X.test,GenoNet.fit)

Ionita-Laza-lab/GenoNet documentation built on Nov. 5, 2019, 2:22 p.m.