PCReg is a user-friendly one-command-line python script kit which can perform linear or non-linear dimensionality reduction and perform principle component regression based on these principle components (PCs) reduced from the linear or non-linear dimensionality reduction. PCReg includes two statistical models for the current version, linear regression and logistic regression. You need to choose the correct model depending on the data type of the response variable (linear regression if the response variable is continuous and logistic regression if the response variable is binary).
PCReg is compiled in Python 3. Please type the following command in your terminal if you have not installed one or some of these python module(s) in python 3:
pip3 install pandas statistics numpy random sklearn scipy statsmodels argparse warnings
You need to provide one matrix file for dimensionality reduction and one covariate matrix file for PC regression. The two matrix files have the same order of individual samples as rows. The dimensionality reduction matrix file has the variables you want to perform dimensionality reduction as the columns. The covariate matrix file has all covariates and a response variable as the columns. Both matrix files are comma-separated. The example matrix files were provided to test the scripts.
The example command to run PCReg:
python PCReg_logistic_regression_argparse_final.py --mtx4pc_file dimensionality_reduction_matrix.csv --out_file res.csv --covars_file covariate_matrix.csv --variance_thre 0.8 --dr_method rbf --covars age sex RIN --dep_var diagnosis
--mtx4pc_file is the argument to provide the dimensionality reduction matrix file, --covars_file is the argument to provide the covariate matrix file, and --out_file is the argument to specify the output file name.
--variance_thre is the argument to provide the threshold for accumulative variance which can be explained by top PCs. The threshold ranges between 0 and 1. Default threshold is 0.8.
--dr_method is the argument to provide the dimensionality reduction method to use. Available options include linear, poly, rbf, sigmoid, cosine. Default method is rbf. This argument is passed from KernelPCA of sklearn. For detailed description please refer to: https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.KernelPCA.html
--covars is the argument to provide covariates you want to include in the model. The covariate names need to match exactly the column names of the covariates in the covariate matrix. If no covariate to provide, please specify it as None (--covars None). NOTE: Please hardcode any character variables into integer variables.
--dep_var is the argument to provide dependent or response variable in the model. The variable name needs to match exactly the column name of the dependent variable in the covariate matrix. NOTE: Please hardcode any character variables into integer variables.
You can get all the argument descriptions with the command:
python PCReg_logistic_regression_argparse_final.py --help
PCReg calculates and exports Chi-square statistics and corresponding P value for the PCs from the PC regression to the file you specified in the argument --out_file.