PRS_PGx_Lasso {PRSPGx}

Construct PGx PRS using penalized regression

Description

Shrink prognostic and predictive effect sizes simultaneously via the penalized term. With different assumptions on the relationship between the two effects, can be PRS-PGx-L (Lasso), PRS-PGx-GL (Group Lasso), and PRS-PGx-SGL (Sparse Group Lasso)

Usage

PRS_PGx_Lasso(Y, Tr, G, intercept = TRUE, lambda, method, alpha = 0.5)

Arguments

Y

a numeric vector containing the quantitative trait

Tr

a numeric vector containing the treatment assignment

G

a numeric matrix containing genotype information

intercept

a logical flag indicating should intercept be fitted (default=TRUE) or set to be FALSE

lambda

a numeric value indicating the penalty

method

a logical flag for different penalized regression methods: 1 = PRS-PGx-L, 2 = PRS-PGx-GL, 3 = PRS-PGx-SGL

alpha

a numeric value indicating the mixing parameter (only used when method = 3). alpha = 1 is the lasso penalty. alpha = 0 is the group lasso penalty

Details

PRS-PGx-Lasso requires individudal-level data

Value

A numeric list, the first sublist contains estimated prognostic effect sizes, the second sublist contains estimated predictive effect sizes

Author(s)

Song Zhai

References

Yang, Y. & Zou, H. A fast unified algorithm for solving group-lasso penalize learning problems. Statistics and Computing 25, 1129-1141 (2015).

Simon, N., Friedman, J., Hastie, T. & Tibshirani, R. Fit a GLM (or cox model) with a combination of lasso and group lasso regularization. R package version, 1 (2015).

Zhai, S., Zhang, H., Mehrotra, D.V. & Shen, J. Paradigm Shift from Disease PRS to PGx PRS for Drug Response Prediction using PRS-PGx Methods (submitted).

Examples

data(PRSPGx.example); attach(PRSPGx.example)
coef_est <- PRS_PGx_Lasso(Y, Tr, G, lambda = 1, method = 1)
summary(coef_est$coef.G)
summary(coef_est$coef.TG)