GENetLib 1.0.4

A Python Library for Gene–environment Interaction Analysis via Deep Learning

GENetLib: A Python Library for Gene–environment Interaction Analysis via Deep Learning

Introduction

GENetLib is a Python library that addresses the lack of portable and friendly software for analyzing gene-environment (G-E) interactions using a deep learning approach with penalization, as developed by Wu et al., 2023. It also tackles the challenge of high-dimensional SNP data analysis by employing a functional data analysis method that reduces data dimensionality, which builds upon with a G-E interaction approach proposed by Ren et al., 2023.

References: Wu, S., Xu, Y., Zhang, Q., & Ma, S. (2023). Gene–environment interaction analysis via deep learning. Genetic Epidemiology, 1–26. https://doi.org/10.1002/gepi.22518 Ren, R., Fang, K., Zhang, Q., & Ma, S. (2023). FunctanSNP: an R package for functional analysis of dense SNP data (with interactions). Bioinformatics, 39(12), btad741. https://doi.org/10.1093/bioinformatics/btad741

Installation

Requirements

matplotlib==3.7.1
numpy==1.24.3
pandas==1.5.3
scipy==1.10.1
setuptools==67.8.0
torch==2.3.0

Normal installation

pip install GENetLib

Mirror

pip install GENetLib -i https://pypi.tuna.tsinghua.edu.cn/simple

Functions

Menu

• SimDataScaler
• SimDataSNP
• ScalerGE
• SNPGE
• GridScalerGE
• GridSNPGE

SimDataScaler

Example data for method ScalerGE and GridScalerGE

Description

Example data for users to apply the method ScalerGE and GridScalerGE.

Usage

SimDataScaler(rho_G, rho_E, dim_G, dim_E, n, dim_E_Sparse = 0, ytype = 'Survival', n_inter = None, linear = True, seed = 0)

Arguments

Arguments	Description
rho_G	numeric, correlation of gene variables.
rho_E	numeric, correlation of environment variables.
dim_G	numeric, dimension of gene variables.
dim_E	numeric, dimension of environment variables.
n	numeric, sample size.
dim_E_Sparse	numeric, dimension of sparse environment variables.
ytype	character, "Survival", "Binary" or "Continuous" type of the output y. If not specified, the default is survival.
n_inter	numeric, number of interaction effect variables.
linear	bool, "True" or "False", whether or not to generate linear data. The default is True.
seed	numeric, random seeds each time when data is generated.

Value

The function "SimDataScaler" outputs a tuple including generated data and the positions of interaction effect variables.

• data: a dataframe contains gene variables, environment variables, interaction variables and output y.
• interaction efecct variables: an array contains the positions of interaction effect variables.

See Also

See also as ScalerGE, GridScalerGE.

Examples

rho_G, rho_E, dim_G, dim_E, n, dim_E_Sparse = 0, ytype = 'Survival', n_inter = None, linear = True, seed = 0
scaler_survival_linear = SimDataScaler(rho_G = 0.25, rho_E = 0.3, dim_G = 500, dim_E = 5, n = 1500, dim_E_Sparse = 2, ytype = 'Survival', n_inter = 30)
scaler_survival_linear_data = scaler_survival_linear[0]
scaler_survival_linear_inter = scaler_survival_linear[1]

SimDataSNP

Example data for method SNPGE and GridSNPGE

Description

Example data for users to apply the method SNPGE and GridSNPGE.

Usage

SimDataSNP(n, m, ytype, seed = 0)

Arguments

Arguments	Description
n	numeric, sample size.
m	numeric, the sequence length of each sample.
ytype	character, "Survival", "Binary" or "Continuous" type of the output y. If not specified, the default is continuous.
seed	numeric, random seeds each time when data is generated.

Value

The function "SimDataScaler" outputs a dictionary including response variable y, scalar variable z and sequence (genotypes) data X.

• x: a matrix representing the sequence data, with the number of rows equal to the number of samples.
• y: an array representing the response variables.
• z: a matrix representing the scalar covariates, with the number of rows equal to the number of samples.
• location: a list defining the sampling sites of the sequence (genotypes) data.

See Also

See also as SNPGE, GridSNPGE.

Examples

snp_continuous = SimDataSNP(n = 1000, m = 100, ytype = 'Continuous', seed = 1)
x = snp_continuous['X']
y = snp_continuous['y']
z = snp_continuous['z']
location = snp_continuous['location']

ScalerGE

G-E interaction analysis via deep leanring when the input X is scaler

Description

This function provides an approach based on deep neural network in conjunction with MCP and L2 penalizations which can simultaneously conduct model estimation and selection of important main G effects and G–E interactions, while uniquely respecting the "main effects, interactions" variable selection hierarchy.

Usage

ScalerGE(data, ytype, dim_G, dim_E, haveGE, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, t = None, model = None, split_type = 0, ratio = [7, 3], important_feature = True, plot = True, model_reg = None, issnp = False)

Arguments

Arguments	Description
data	dataframe or list, follow the format: dataframe with {G, GE(optional), E, y} or list with {y, G, E, GE(optional)}.
ytype	character, "Survival", "Binary" or "Continuous" type of the output y.
dim_G	numeric, dimension of gene variables.
dim_E	numeric, dimension of environment variables.
haveGE	bool, "True" or "False", whether there are GE interactions in the data. If not, the function will calculate GE interactions.
num_hidden_layers	numeric, number of hidden layers in the neural network.
nodes_hidden_layer	list, contains number of nodes in each hidden layer.
Learning_Rate2	numeric, learning rate of hidden layers.
L2	numeric, tuning parameter of L2 penalization.
Learning_Rate1	numeric, learning rate of sparse layers.
L	numeric, tuning parameter of MCP penalization.
Num_Epochs	numeric, number of epochs for neural network training.
t	numeric, threshold in the selection ofimportant features.
model	tuple, pre-trained models. If not specified, the default is none.
split_type	integer, types of data split. If split_type = 0, the data is divided into a training set and a validation set. If split_type = 1, the data is divided into a training set, a validation set and a test set.
ratio	list, the ratio of data split.
important_feature	bool, "True" or "False", whether or not to show output features.
plot	bool, "True" or "False", whether or not to show the line plot of residuals with the number of neural network epochs.

Value

The function "ScalerGE" outputs a tuple including training results of the neural network.

• Residual of the training set.
• Residual of the validation set.
• C index(y is survival) or R2(y is continuous or binary) of the training set.
• C index(y is survival) or R2(y is continuous or binary) of the validation set.
• A neural network after training.
• Important features of gene variables.
• Important features of G-E interaction variables.

See Also

See also as SimDataScaler, GridScalerGE.

Examples

from SimDataScaler import SimDataScaler
ytype = 'Survival'
num_hidden_layers = 2
nodes_hidden_layer = [1000, 100]
Learning_Rate2 = 0.035
L2 = 0.01
Learning_Rate1 = 0.06
L = 0.09
Num_Epochs = 100
t = 0.01
scaler_survival_linear = SimDataScaler(rho_G = 0.25, rho_E = 0.3, dim_G = 500, dim_E = 5, n = 1500, dim_E_Sparse = 2, ytype = 'Survival', n_inter = 30)
ScalerGERes = ScalerGE(scaler_survival_linear[0], ytype, 500, 5, True, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, t, split_type = 0, ratio = [7, 3], important_feature = True, plot = True)

SNPGE

G-E interaction analysis via deep leanring when the input X is SNP

Description

This function provides an approach based on deep neural network in conjunction with MCP and L2 penalizations, which treats dense SNP measurements as a realization of a genetic function and can "bypass" the dimensionality challenge.

Usage

SNPGE(y, z, location, X, ytype, btype, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, nbasis1, params1, t = None, Bsplines = 20, norder1 = 4, model = None, split_type = 0, ratio = [7, 3], plot_res = True, plot_beta = True)

y|numeric, an array representing the response variables. z|numeric, a matrix representing the scalar covariates, with the number of rows equal to the number of samples. location|list, a list defining the sampling sites of the sequence (genotypes) data. X|numeric, a matrix representing the sequence data, with the number of rows equal to the number of samples. ytype|character, "Survival", "Binary" or "Continuous" type of the output y. btype|character, "Bspline", "Exponential", "Fourier", "Monomial" or "power" type of spline. num_hidden_layers|numeric, number of hidden layers in the neural network. nodes_hidden_layer|list, contains number of nodes in each hidden layer. Learning_Rate2|numeric, learning rate of hidden layers. L2|numeric, tuning parameter of L2 penalization. Learning_Rate1|numeric, learning rate of sparse layers. L|numeric, tuning parameter of MCP penalization. Num_Epochs|numeric, number of epochs for neural network training. nbasis1|integer, an integer specifying the number of basis functions that constitutes the genetic variation function. params1|integer, in addition to rangeval1 (a vector of length 2 giving the lower and upper limits of the range of permissible values for the genetic variation function) and nbasis1, all bases have one or two parameters unique to that basis type or shared with one other. Bsplines|integer, an integer specifying the number of basis functions that constitutes the genetic effect function. norder1|integer, an integer specifying the order of bsplines that constitutes the genetic effect function, which is one higher than their degree. The default of 4 gives cubic splines. model|tuple, pre-trained models. If not specified, the default is none. split_type|integer, types of data split. If split_type = 0, the data is divided into a training set and a validation set. If split_type = 1, the data is divided into a training set, a validation set and a test set. ratio|list, the ratio of data split. plot_res|bool, "True" or "False", whether or not to show the line plot of residuals with the number of neural network epochs. plot_beta|bool, "True" or "False", whether or not to show the graph of predicted functions.

Value

The function "SNPGE" outputs a tuple including training results of the neural network.

• Residual of the training set.
• Residual of the validation set.
• C index(y is survival) or R2(y is continuous or binary) of the training set.
• C index(y is survival) or R2(y is continuous or binary) of the validation set.
• A neural network after training.
• Estimated coefficients of the chosen basis functions for the genetic effect function beta0(t) and interaction items betak(t).
• The estimated genetic effect function beta(t) and interaction items betak(t).

See Also

See also as SimDataSNP, GridSNPGE.

Examples

from SimDataSNP import SimDataSNP
num_hidden_layers = 2
nodes_hidden_layer = [100,10]
Learning_Rate2 = 0.035
L2 = 0.01
Learning_Rate1 = 0.02
L = 0.01
Num_Epochs = 50
nbasis1 = 5
params1 = 4
seed = 123
snp_continuous = SimDataSNP(n = 1500, m = 30, ytype = 'Continuous', seed)
y = snp_continuous['y']
z = snp_continuous['z']
location = snp_continuous['location']
X = snp_continuous['X']
SNPGE_Res = SNPGE(y, z, location, X, 'Continuous', 'Bspline', num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, nbasis1, params1, Bsplines = 5, norder1 = 4, model = None, split_type = 1, ratio = [3, 1, 1], plot_res = True)

GridScalerGE

Grid search for ScalerGE

Description

This function performs grid search for ScalerGE over a grid of values for the regularization parameter L, L2 and learning rate Learning_Rate1, Learning_Rate2.

Usage

GridScalerGE(data, ytype, dim_G, dim_E, haveGE, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, t = None, model = None, split_type = 0, ratio = [7, 3], important_feature = True, plot = True, model_reg = None, issnp = False)

Arguments

Arguments	Description
data	dataframe or list, follow the format: dataframe with {G, GE(optional), E, y} or list with {y, G, E, GE(optional)}.
ytype	character, "Survival", "Binary" or "Continuous" type of the output y.
dim_G	numeric, dimension of gene variables.
dim_E	numeric, dimension of environment variables.
haveGE	bool, "True" or "False", whether there are GE interactions in the data. If not, the function will calculate GE interactions.
num_hidden_layers	numeric, number of hidden layers in the neural network.
nodes_hidden_layer	list, contains number of nodes in each hidden layer.
Learning_Rate2	list, learning rate of hidden layers.
L2	list, tuning parameter of L2 penalization.
Learning_Rate1	list, learning rate of sparse layers.
L	list, tuning parameter of MCP penalization.
Num_Epochs	numeric, number of epochs for neural network training.
t	numeric, threshold in the selection ofimportant features.
model	tuple, pre-trained models. If not specified, the default is none.
split_type	integer, types of data split. If split_type = 0, the data is divided into a training set and a validation set. If split_type = 1, the data is divided into a training set, a validation set and a test set.
ratio	list, the ratio of data split.
important_feature	bool, "True" or "False", whether or not to show output features.
plot	bool, "True" or "False", whether or not to show the line plot of residuals with the number of neural network epochs.

Value

The function "GridScalerGE" outputs a tuple including training results of the neural network.

• Values of tunning parameters after grid search.
• Residual of the training set.
• Residual of the validation set.
• C index(y is survival) or R2(y is continuous or binary) of the training set.
• C index(y is survival) or R2(y is continuous or binary) of the validation set.
• A neural network after training.
• Important features of gene variables.
• Important features of GE interaction variables.

See Also

See also as SimDataScaler, ScalerGE.

Examples

from SimDataScaler import SimDataScaler
ytype = 'Survival'
num_hidden_layers = 2
nodes_hidden_layer = [1000, 100]
Learning_Rate2 = [0.035, 0.045]
L2 = [0.1]
Learning_Rate1 = [0.01, 0.02, 0.03, 0.04, 0.05, 0.06]
L = [0.04, 0.05, 0.06, 0.07, 0.08, 0.09]
Num_Epochs = 100
t = 0.01
dim_E = 5
dim_G = 500
haveGE = True
scaler_survival_linear = SimDataScaler(rho_G = 0.25, rho_E = 0.3, dim_G = 500, dim_E = 5, n = 1500, dim_E_Sparse = 2, ytype = 'Survival', n_inter = 30)
GridScalerGERes = GridScalerGE(scaler_survival_linear[0], ytype, dim_G, dim_E, haveGE, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, t, split_type = 1, ratio = [3, 1, 1], plot = True)

GridSNPGE

Grid search for SNPGE

Description

This function performs grid search for SNPGE over a grid of values for the regularization parameter L, L2 and learning rate Learning_Rate1, Learning_Rate2.

Usage

GridSNPGE(y, z, location, X, ytype, btype, num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, nbasis1, params1, t = None, Bsplines = 20, norder1 = 4, model = None, split_type = 0, ratio = [7, 3], plot_res = True, plot_beta = True)

y|numeric, an array representing the response variables. z|numeric, a matrix representing the scalar covariates, with the number of rows equal to the number of samples. location|list, a list defining the sampling sites of the sequence (genotypes) data. X|numeric, a matrix representing the sequence data, with the number of rows equal to the number of samples. ytype|character, "Survival", "Binary" or "Continuous" type of the output y. btype|character, "Bspline", "Exponential", "Fourier", "Monomial" or "power" type of spline. num_hidden_layers|numeric, number of hidden layers in the neural network. nodes_hidden_layer|list, contains number of nodes in each hidden layer. Learning_Rate2|list, learning rate of hidden layers. L2|list, tuning parameter of L2 penalization. Learning_Rate1|list, learning rate of sparse layers. L|list, tuning parameter of MCP penalization. Num_Epochs|numeric, number of epochs for neural network training. nbasis1|integer, an integer specifying the number of basis functions that constitutes the genetic variation function. params1|integer, in addition to rangeval1 (a vector of length 2 giving the lower and upper limits of the range of permissible values for the genetic variation function) and nbasis1, all bases have one or two parameters unique to that basis type or shared with one other. Bsplines|integer, an integer specifying the number of basis functions that constitutes the genetic effect function. norder1|integer, an integer specifying the order of bsplines that constitutes the genetic effect function, which is one higher than their degree. The default of 4 gives cubic splines. model|tuple, pre-trained models. If not specified, the default is none. split_type|integer, types of data split. If split_type = 0, the data is divided into a training set and a validation set. If split_type = 1, the data is divided into a training set, a validation set and a test set. ratio|list, the ratio of data split. plot_res|bool, "True" or "False", whether or not to show the line plot of residuals with the number of neural network epochs. plot_beta|bool, "True" or "False", whether or not to show the graph of predicted functions.

Value

The function "GridSNPGE" outputs a tuple including training results of the neural network.

• Values of tunning parameters after grid search.
• Residual of the training set.
• Residual of the validation set.
• C index(y is survival) or R2(y is continuous or binary) of the training set.
• C index(y is survival) or R2(y is continuous or binary) of the validation set.
• A neural network after training.
• Estimated coefficients of the chosen basis functions for the genetic effect function beta0(t) and interaction items betak(t).
• The estimated genetic effect function beta(t) and interaction items betak(t).

See Also

See also as SimDataSNP, SNPGE.

Examples

from SimDataSNP import SimDataSNP
num_hidden_layers = 2
nodes_hidden_layer = [100, 10]
Learning_Rate2 = [0.005, 0.01, 0.015]
L2 = [0.005, 0.01, 0.015]
Learning_Rate1 = [0.001, 0.005]
L = [0.005, 0.006, 0.007]
Num_Epochs = 50
nbasis1 = 5
params1 = 4
snp_continuous = SimDataSNP(n = 1000, m = 30, 'Continuous', seed = 1)
y = snp_continuous['y']
z = snp_continuous['z']
location = snp_continuous['location']
X = snp_continuous['X']
GridSNPGE_Res = GridSNPGE(y, z, location, X, 'Continuous', 'Bspline', num_hidden_layers, nodes_hidden_layer, Learning_Rate2, L2, Learning_Rate1, L, Num_Epochs, nbasis1, params1, Bsplines = 5, norder1 = 4, model = None, split_type = 0, ratio = [7,3], plot_res = True)