GENetLib: A Python Library for Gene–environment Interaction Analysis via Deep Learning
GENetLib is a Python library designed for gene-environment interaction analysis via neural network, addressing the analytical challenges in complex disease research.
This package is available at GitHub, and is capable of handling a variety of input data types:
- Scalar input data
- Functional input data (or densely measured data)
This package also supports diverse output requirements:
- Continuous output data
- Binary output data
- Survival output data
By integrating minimax concave penalty (MCP) and L2-norm regularization within a neural network estimation framework, GENetLib offers an innovative solution for high-dimensional genetic data analysis. The framework is shown below.
We provide a web-based documentation which introduces the meaning of function parameters, the usage of functions, detailed information about methods, and gives examples for each. The web page is available at documentations. This package has been uploaded to PyPI with previous versions, and the web page is available at PyPI package. Users can also check releases to get historical versions.
Features
GENetLib has the following features:
- Comprehensiveness: Supports a variety of input and output formats, enabling the construction of comprehensive neural network models for G-E interaction analysis.
- Flexibility: Offers a multitude of parameters allowing users to build models flexibly according to their specific needs.
- Functional data compatibility: Implements methods for functional data analysis (FDA) in Python, facilitating the processing of functional data with Python.
- Scalability: New methods for G-E interaction analysis via deep learning can be easily integrated into the system.
Installation
It is recommended to use pip for installation:
pip install GENetLib
To get further information about installation and independencies, please move to installation instructions.
Quick Start
We start with the two basic functions scalar_ge and func_ge.
scalar_ge
scalar_ge performs G-E interaction analysis via deep leanring when the input is scalar data.
from GENetLib.sim_data_scalar import sim_data_scalar
from GENetLib.scalar_ge import scalar_ge
# Get example data where input is scalar data and output is survival data
scalar_survival_linear = sim_data_scalar(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)
# Set up the ScalerGE model
scalar_ge_res = scalar_ge(y = scalar_survival_linear['y'], G = scalar_survival_linear['G'], E = scalar_survival_linear['E'],
ytype = 'Survival',num_hidden_layers = 2, nodes_hidden_layer = [1000, 100], num_epochs = 100,
learning_rate1 = 0.06, learning_rate2 = 0.035, lambda1 = None, lambda2 = 0.09, Lambda = 0.1,
threshold = 0.01, split_type = 0, ratio = [7, 3], important_feature = True, plot = True)
func_ge
func_ge performs G-E interaction analysis via deep leanring when the input is functional data.
from GENetLib.sim_data_func import sim_data_func
from GENetLib.func_ge import func_ge
# Get example data where input is densely measured functional data and output is survival data
func_continuous = sim_data_func(n = 1500, m = 30, ytype = 'Continuous', seed = 123)
# Set up the FuncGE model
func_ge_res = func_ge(y = func_continuous['y'], X = func_continuous['X'], location = func_continuous['location'],
Z = func_continuous['Z'], ytype = 'Continuous', btype = 'Bspline', num_hidden_layers = 2,
nodes_hidden_layer = [100,10], num_epochs = 50, learning_rate1 = 0.02, learning_rate2 = 0.035,
nbasis1 = 5, params1 = 4, lambda1 = None, lambda2 = 0.01, Lambda = 0.01, Bsplines = 5,
norder1 = 4, split_type = 1, ratio = [3, 1, 1], plot_res = True)
For more information about the functions and methods, please check main functions.
License
GENetLib is licensed under the MIT License. See LICENSE for details.
Feedback
If you are interested in this, please check GitHub. Thanks for supports!
