data/Correlation/correlation.py

# Import
import cProfile
from distutils import core
from pathlib import Path
import numpy as np
import scipy as sp
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
import seaborn as sns
import os
import pickle
import scipy.stats as stats
import argparse
from adjustText import adjust_text
import matplotlib.backends.backend_pdf



def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('-b', type = str, help = 'Path to barcodes')
    parser.add_argument('-xp', type = str, help = 'Experiment ID')
    parser.add_argument('-c', type = str, default = "D:\BCCAncer\FILES\codebook_0_C1E1_van.csv", help = 'Path to codebook')
    parser.add_argument('-a', type = str, default = "D:\BCCAncer\FILES\XP2474_4T1_C1E1_new_bulk.csv", help = 'Path to bulk file (abundance file)')
    parser.add_argument('-o', type = str, default = "D:\BCCAncer\EXP\XP7174\Correlations\BulkCorr", help = 'Path to save processed files')
    parser.add_argument('-is_cambridge', type = bool, default = False, help = 'Are the results for Cambridge datasets?')
    parser.add_argument('-drop_blanks', type = bool, default = False, help = 'Remove blanks before correlation')
    parser.add_argument('-log', type =bool, default = True, help = 'Do we want to take log while correlation of bulk?')
    parser.add_argument("-d", nargs="+", type = float, default=0.65, help = 'Max mean distance threshold for an area of detected barcode')
    parser.add_argument("-removeZ", nargs="+", type=int, default=None, help="Z slices to remove from evaluation")
    #parser.add_argument('-gene_list', type = str, action='append', required=True, help = 'List of genes which we want to exclude from analysis')


    args = parser.parse_args()

    if not os.path.exists(args.o):
        os.makedirs(args.o)

    info_np = np.zeros(shape = (len(args.d), 7))

    # read sheet and remove undesired z slices
    df = read_sheet(args.b)
    if args.removeZ is not None:
        removeZ_set = set(args.removeZ)
        df = df[~df['z'].isin(removeZ_set)]
    
    
    
    pdf = matplotlib.backends.backend_pdf.PdfPages(f'{args.o}/{args.xp}_correlation.pdf')
    correlation = Correlation(args.c, df, args.a, args.o, args.xp, args.is_cambridge, pdf)
    args.d.sort(reverse = True)

    
    for idx, dist_threshold in enumerate(args.d):
        filter_df = correlation.filter_distance(dist_threshold)
        gp_df = correlation.groupby(filter_df)
        gp_df = correlation.merge_df_cb(gp_df)

        tot_counts = gp_df.counts.sum()
        blank_counts = gp_df.loc[gp_df['gene_symbol'].isin(['Blank_01', 'Blank_02', 'Blank_03', 'Blank_04', 'Blank_05', 'Blank_06'])].counts.sum()

        info_np[idx][0] = dist_threshold
        info_np[idx][3] = int(tot_counts)
        info_np[idx][4] = int(blank_counts)
        info_np[idx][5] = int(len(gp_df))

        gp_df = correlation.df_bulk(gp_df)

        if args.drop_blanks:
            gp_df = correlation.remove_blanks(gp_df) 

        """
            if args.gene_list:
            print('removing')
            gp_df = correlation.remove_specific_genes(gp_df)
        """

        info_np[idx][-1] = len(gp_df)

        gp_df['log_counts'] = np.log2(gp_df['counts']+1)
        gp_df['log_tpm'] = np.log2(gp_df['bulk_exp']+0.0001)

        gp_df.to_csv(f'{args.o}/{args.xp}_{dist_threshold}.csv')


        info_np[idx][1], info_np[idx][2] = correlation.log_correlation(gp_df, dist_threshold)

    pdf.close()

    info_df = pd.DataFrame(info_np, columns = ['Distance Threshold',\
                                        'Pearson correlation',\
                                        'Spearman correlation',\
                                        '# detected barcodes (including control barcodes)',\
                                        '# detected control (blanks) barcodes',\
                                        '# genes at dist threshold in correlation',\
                                        '# barcodes counted towards correlaton estimation'])
    info_df.to_csv(f'{args.o}/info_{args.xp}.csv')
 
def read_sheet(file):
    
    ext = file.split(".")[-1]
    if ext == "csv":
        df = pd.read_csv(file)
    elif ext == "tsv":
        df = pd.read_csv(file, '\t')
    elif ext in {"xls", "xlsx", "xlsm", "xlsb"}:
        df = pd.read_excel(file)
    else:
        raise ValueError("Unexpected file extension")
    return df    

class Correlation:
    def __init__(self, codebook, barcodes, bulk, output_dir, name, is_cambridge, pdf_object):
        self.codebook = read_sheet(codebook)
        self.barcodes = barcodes
        self.bulk = read_sheet(bulk)
        self.output_dir = output_dir
        self.name = name
        self.is_cambridge = is_cambridge
        self.pdf_object = pdf_object
        #self.remove_genes_list = remove_genes_list

    def filter_distance(self, distance_threshold):
        return self.barcodes.loc[self.barcodes['mean_distance']<distance_threshold]

    def groupby(self, df):
        return (df.groupby('barcode_id').size().reset_index(name='counts'))

    def modify_codebook(self):
        self.codebook['barcode_id'] = self.codebook.index
        self.codebook.rename(columns={"name": "gene_symbol"}, inplace = True)

    def merge_df_cb(self,df):
        self.modify_codebook()
        return pd.merge(self.codebook, df, how='inner', on='barcode_id')

    def df_bulk(self, df):
        return pd.merge(self.bulk, df, how='inner', on='gene_symbol')

    def remove_blanks(self,df):
        return df[df['bulk_exp'] != 0]

    """
    def remove_specific_genes(self,df):
        c = 0
        print(self.remove_genes_list)
        for gene in self.remove_genes_list:
            print(c, gene)
            df = df.loc[df['gene_symbol']!=gene]
            #cccprint(df.loc[df['gene_symbol']==gene])
            c+=1
        return df
    """
    

    def log_correlation(self, df, dist_threshold):
            f1, ax = plt.subplots(figsize=(9, 9))
            sns.set_palette("deep")
            sns.scatterplot(x="log_tpm",y="log_counts",data=df,ax=ax)
                           
            if self.is_cambridge == True:
                count_type = "C_counts"
            else: count_type = "V_counts"
                    
                
            ax.set_title("TPM Correlation for " + self.name, fontsize = 15)
            ax.set_xlabel("log2(TPM+1e-4), V_bulk", fontsize = 20)
            ax.set_ylabel("log2(# detected counts+1), "+ count_type, fontsize = 20)
            
            
            def plotlabel(xvar, yvar, label):
                ax.text(xvar+0.02, yvar, label)
      
                
            pearson, _ = stats.pearsonr(df["log_tpm"],df["log_counts"])
            spearman, _ = stats.spearmanr(df["log_tpm"],df["log_counts"])
            #kendalltau, _ = stats.kendalltau(df["log_tpm"],df["log_counts"])
               
            ax.text(.01, .95, 'Pearson = {:.2f}\nSpearman = {:.2f}'.format(pearson,spearman),transform=ax.transAxes)
            #df.apply(lambda x: plotlabel(x['log_tpm'],  x['log_counts'], x['gene_symbol']), axis=1)
            #plt.axvline(x = 0, color = 'r', label = 'axvline - full height')
            
            self.pdf_object.savefig(f1)

            texts = []
            for xs,ys,label in zip(df['log_tpm'],df['log_counts'],df['gene_symbol']):
                    texts.append(ax.text(xs,ys,label))
            adjust_text(texts, force_points=0.2, force_text=0.2,expand_points=(1, 1), expand_text=(1, 1), arrowprops=dict(arrowstyle="-", color='black', lw=0.5))
            plt.tight_layout()
            
            #plt.savefig(f'{self.output_dir}/{self.name}_{dist_threshold}.pdf')
            self.pdf_object.savefig(f1)

            return pearson, spearman


if __name__ == '__main__':
    #cProfile.run('main()')
    main()