main.py

import pandas as pd
import argparse
import data_processing as d
import reg_main as r
import post_processing as p
import pdb


def highdim_sldsc(args):
    """ This function 
        1. process user input
        and return a data object with everything needed for regression analysis. 
        See data_processing.py for attributes of data object.
        2. perform regression on the processed data
        and return a regression object carrying learned coefficients and bias.
        See regression.py for details.
        3. post process the learned coefficients and bias. 
        Use the un-standardized coefficients and bias for prediction on test set.
        Save the un-standardized coefficients and bias to file
    """
    d.match_SNPs(args)
    num_SNPs = d.get_num_SNPs(args)
    original_data = d.data(args.ld,args.sumstats,args.weights_ld,[[0,num_SNPs]])
    # TODO add shuffling data step
    train_ind,test_ind = d.get_traintest_ind(args)
    # train_ind and test_ind are lists of lists of two elements containing end points TODO change code related to active_ind
    weights_fname = d.weights_processing(args,original_data) # compute final weights and store in file. Final weights are the inverse of the estimated variances of the noise #TODO changed final weights from inv_sqrt of variances into inv of variances, need to change related code
    # form train and test data objects
    train_data = d.data(args.ld,args.sumstats,weights_fname,train_ind)
    test_data = d.data(args.ld,args.sumstats,weights_fname,test_ind)
    #perform regression with specified parameters on training data
    reg = r.perform_regression(args,train_data)
    p.process(args,reg,test_data) # postprocessing on testing data
    return



if __name__=='__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--highdim-sldsc',action='store_true',help='run high dimensional stratified ldscore regression')
    parser.add_argument('--ld',help='The .h5 file combining ldscores for all annotations')
    parser.add_argument('--sumstats',help='Text file of formatted summary statistics with columns SNP, A1, A2, N, CHISQ, Z, where CHISQ is Z^2')
    parser.add_argument('--weights_ld',help='The user input weights. Chrom separated "weights.#chr.l2.ldscore.gz" files input as --weights weights.')
    parser.add_argument('--annot-snplist',help='The text file listing all SNPs used to create annotation ld. This file should contain two columns CHR and SNP. The order matches the order of SNPs used in --ld.')
    parser.add_argument('--leave-out',help='Specify the region to leave out, for example, chr22.')
    parser.add_argument('--reg-method',help='Name of the method to run regression with. Choose among OLS,Lasso,Lasso+OLS,elnet,grpLasso,skLassoCV,skOLS')
    parser.add_argument('--output-folder',help='Point to a location where the program can store output files')
    parser.add_argument('--fit',help='the method of fitting, this is usually for testing. Available values: direct--meaning using sklearn prepackaged methods to fit, no manual weighting or scaling; manual--manually scale and weight the data, then recover accordingly.')
    args = parser.parse_args()
    
    if args.highdim_sldsc:
        highdim_sldsc(args)