offtracker 1.0.2__zip → 2.7.7__zip

offtracker-1.0.2/scripts/offtracker_analysis.py

@@ -1,407 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-
-import os,glob,sys,re,time,shutil
-
-if sys.version_info < (3,0):
-    import platform
-    raise Exception(f'python3 is needed, while running {platform.python_version()} now')
-
-import offtracker
-from offtracker.X_analysis import *
-script_dir = os.path.abspath(os.path.dirname(offtracker.__file__))
-script_folder= os.path.join(script_dir, 'mapping')
-
-import argparse
-import pandas as pd
-import numpy as np
-import pybedtools
-from scipy.stats import norm
-
-def main():
-    parser = argparse.ArgumentParser()
-    parser.description='Analyze the ssChIP-seq data.'
-    parser.add_argument('-f','--folder' , type=str, required=True,  nargs='+', help='Directory of the data folder.' )
-    parser.add_argument('--seqfolder'   , type=str, required=True,  help='Directory of sgRNA information created by seq_cadidates.')
-    parser.add_argument('--name'        , type=str, required=True,  help='custom name of the sgRNA' )
-    parser.add_argument('--exp'         , type=str, default='all',  nargs='+', help='A substring mark in the name of experimental samples. The default is to use all samples other than control' )
-    parser.add_argument('--control'     , type=str, default='none', nargs='+', help='A substring mark in the name of control samples. The default is no control. "others" for all samples other than --exp.' )
-    parser.add_argument('--regions'     , type=str, default='auto', nargs='+', help='Regions around candidate sites.' )
-    parser.add_argument('--rgO'         , type=str, default='mean', help='Regoins operation. mean/min/max for regions' )
-    parser.add_argument('--repO'        , type=str, default='mean', help='Replicate operation for multiple experimental samples (mean/min/max)' )
-    parser.add_argument('--repOC'       , type=str, default='mean', help='Replicate operation for multiple control samples (mean/max)' )
-    parser.add_argument('-g','--genome' , type=str, default='hg38', help='File of chromosome sizes, or "hg38", "mm10" ')
-    parser.add_argument('-o','--outdir' , type=str, default='first',help='The output folder. Default is the first folder of --folder' )
-    parser.add_argument('--outname'     , type=str, default='same', help='The suffix of output files. Default is the same --exp' )
-    parser.add_argument('--overwrite'   , action='store_true', help='Whether to overwrite existed dataframes.' )
-    parser.add_argument('--NoShapeBonus', action='store_true', help='Disable shape bonus for each site.' )
-    parser.add_argument('--SignalFormula',type=int, default=4, help='The signal formula' )
-    parser.add_argument('--ScoreFormula', type=int, default=3, help='The score formula' )
-    parser.add_argument('--ShapeFormula', type=int, default=2, help='The shape formula' )
-    parser.add_argument('--ShapeThresh',  type=str, default='auto', help='The threshold for shape formula' )
-    parser.add_argument('--ShapeWeight',  type=str, default='auto', help='The weight for shape bonus' )
-    parser.add_argument('--SeqScorePower',type=float, default=2, help='The seq score power' )
-    parser.add_argument('--ScorePseudocount',type=float, default=0.9, help='The score pseudocount' )
-    parser.add_argument('--RescueFactor',type=float, default=1.2, help='The rescue factor' )
-    parser.add_argument('--RescueThresh1',type=float, default=1.5, help='The rescue threshold for min raw' )
-    parser.add_argument('--RescueThresh2',type=float, default=1.8, help='The rescue threshold for raw score' )
-    parser.add_argument('--clean', action='store_true', help='Whether to remove temp files')
-    
-    print(f'Runing offtracker verision: {offtracker.__version__}')
-    # main parameters
-    args = parser.parse_args()
-    folders = args.folder
-    seq_folder = args.seqfolder
-    sgRNA_name = args.name
-    control_mark = args.control
-    exp_mark = args.exp
-    ctr_samples = []
-    outdir = args.outdir
-    if outdir == 'first':
-        outdir = folders[0]
-    os.chdir(outdir)
-    for a_folder in folders:
-        temp_dir = os.path.join(a_folder, 'temp')
-        if not os.path.exists( temp_dir ):
-            os.makedirs(temp_dir)
-    
-    # parameters for algorithm
-    operator = 'p'
-    
-    dict_signal_formula = {1:signal_formula1, 2:signal_formula2, 3:signal_formula3, 4:signal_formula4}
-    dict_score_formula = {1:score_formula1, 2:score_formula2, 3:score_formula3, 4:score_formula4, 5:score_formula5}
-    dict_shape_formula = {1:shape_formula1, 2:shape_formula2}
-
-    the_signal_formula = dict_signal_formula[args.SignalFormula]
-    if control_mark == 'none':
-        print('No control samples, all samples in the folder are regarded as experimental samples.')
-        the_score_formula = score_formula5
-    else:
-        the_score_formula = dict_score_formula[args.ScoreFormula]
-    if args.NoShapeBonus:
-        the_shape_formula = None
-        print('the_shape_formula is None')
-    else:
-        the_shape_formula = dict_shape_formula[args.ShapeFormula]
-    
-    if args.ShapeFormula == 1:
-        ratio_thresh = 1.5
-        exp_weight = 0.3
-    elif args.ShapeFormula == 2:
-        ratio_thresh = 1
-        exp_weight = 0.5
-    
-    if args.ShapeThresh != 'auto':
-        ratio_thresh = float(args.ShapeThresh)
-    if args.ShapeWeight != 'auto':
-        exp_weight = float(args.ShapeWeight)
-    
-    seq_score_power = args.SeqScorePower
-    if args.regions == 'auto':
-        regions = [500, 1000, 2000, 3000]
-    else:
-        regions = list(map(int, args.regions))
-    print('regions:',regions)
-    bkgs = ['bkg1','bkg2','bkg3']
-    noise_length = 5000
-    max_region = max(regions)
-    region_op = args.rgO
-    repO = args.repO
-    repOC = args.repOC
-    score_pseudo = args.ScorePseudocount
-    rescue_factor = args.RescueFactor
-    assert repO in ['mean','min','max'], "--repO only accepts \"mean\",\"min\",or \"max\" "
-    assert repOC in ['mean','max'], "--repO only accepts \"mean\" or \"max\" "
-    print('repO',repO)
-    print('repOC',repOC)
-    print('ShapeThresh:',ratio_thresh)
-    print('ShapeWeight:',exp_weight)
-    print('SignalFormula:',args.SignalFormula)
-    print('ScoreFormula:',args.ScoreFormula)
-    print('ShapeFormula:',args.ShapeFormula)
-    print('SeqScorePower:',args.SeqScorePower)
-    print('ScorePseudocount:',args.ScorePseudocount)
-    print('RescueFactor:',args.RescueFactor)
-    # glob samples, check paired
-    all_samples_name = []
-    all_samples_pref = []
-    for a_folder in folders:    
-        forward_bed_files = glob.glob( os.path.join( a_folder, '*.fw.bed' ) )
-        forward_bed_files.sort()
-        reverse_bed_files = glob.glob( os.path.join( a_folder, '*.rv.bed' ) )
-        reverse_bed_files.sort()
-        forward_samples = pd.Series(forward_bed_files).apply(os.path.basename).str[:-7]
-        reverse_samples = pd.Series(reverse_bed_files).apply(os.path.basename).str[:-7]
-        assert (forward_samples == reverse_samples).all()
-        all_samples_name.extend( forward_samples )
-        all_samples_pref.extend( pd.Series(forward_bed_files).str[:-7] )
-    all_samples_pref = pd.Series(all_samples_pref)
-    all_samples_name = pd.Series(all_samples_name)
-
-    if control_mark == 'none':
-        if exp_mark == 'all':
-            exp_samples = list( all_samples_name )
-        else:
-            exp_samples = []
-            for a_mark in exp_mark:
-                exp_samples.extend( list( all_samples_name[all_samples_name.str.contains(a_mark)] ) )
-    elif control_mark == 'others':
-        if exp_mark == 'all':
-            exp_samples = list( all_samples_name )
-        else:
-            exp_samples = []
-            for a_mark in exp_mark:
-                exp_samples.extend( list( all_samples_name[all_samples_name.str.contains(a_mark)] ) )
-            ctr_samples = list( all_samples_name[~all_samples_name.isin(exp_samples)] )
-    else:
-        for a_mark in control_mark:
-            ctr_samples.extend( list( all_samples_name[all_samples_name.str.contains(a_mark)] ) )
-        if exp_mark == 'all':
-            exp_samples = list( all_samples_name[~all_samples_name.isin(ctr_samples)] )
-        else:
-            exp_samples = []
-            for a_mark in exp_mark:
-                exp_samples.extend( list( all_samples_name[all_samples_name.str.contains(a_mark)] ) )
-    n_exp = len(exp_samples)
-    n_ctr = len(ctr_samples)
-    print(f'Experimental group has {n_exp} samples:\n{exp_samples}')
-    print(f'Control group has {n_ctr} samples:\n{ctr_samples}')
-
-    # mark 错误时
-    assert n_exp > 0, 'No experimental sample is found. Please check the name pattern.'
-    if (n_ctr==0)&(control_mark != 'none'):
-        print('Name pattern for control sample(s) was given, but no file meet the pattern.')
-        return 'Program terminated'
-    
-    # sequence score
-    try:
-        df_alignment = pd.read_csv(os.path.join(seq_folder, f'df_alignment_{sgRNA_name}_{max_region}.csv'), index_col=0)
-    except FileNotFoundError:
-        return 'Please run offtracker_candidates.py first and provide the correct directory with --seqfolder'
-    
-    # chromosome sizes
-    if (args.genome == 'hg38') or (args.genome == 'mm10'):
-        dir_chrom_sizes = os.path.join(script_folder,f'{args.genome}.chrom.sizes')
-    else:
-        dir_chrom_sizes = args.genome
-    
-    # sgRNA length
-    sgRNA_fa = os.path.join(seq_folder, sgRNA_name + '_PAM.fasta')
-    if not os.path.isfile(sgRNA_fa):
-        raise Exception(f'{sgRNA_name}_PAM.fasta is not in the seqfolder')
-    with open(sgRNA_fa,'r') as f:
-        temp = f.readlines()
-        len_sgRNA_PAM = len(temp[1].strip())
-    #
-    selected_samples_pref = all_samples_pref[all_samples_name.isin(exp_samples+ctr_samples)]
-    selected_samples_name = all_samples_name[all_samples_name.isin(exp_samples+ctr_samples)]
-    # read counts on candidate regions
-    for a_pref in selected_samples_pref:
-        cand_count(a_pref, sgRNA_name, regions, seq_folder, dir_chrom_sizes, overwrite=False)
-        bkg_count(a_pref, sgRNA_name, bkgs, seq_folder, dir_chrom_sizes, overwrite=False)
-    
-
-    ############
-    ## signal ##
-    ############
-    if args.outname == 'same':
-        outname = exp_mark
-    else:
-        outname = args.outname
-    output = f'./temp/df_pivot_{outname}.csv'
-    if (os.path.isfile(output))&(not args.overwrite):
-        print(f'skip {output}')
-    else:
-        # 计算每个样本的 df_counts
-        dict_df_counts = {}
-        for a_pref, a_sample in zip(selected_samples_pref, selected_samples_name):
-            dirname = os.path.dirname(a_pref)
-            basename = os.path.basename(a_pref)
-            temp_dir = os.path.join(dirname, 'temp')
-            a_pref = os.path.join(temp_dir, basename)
-            dir_df_counts = f'{a_pref}_{sgRNA_name}_df_counts_{the_signal_formula.__name__}_{operator}.csv'
-            if (os.path.isfile(dir_df_counts))&(not args.overwrite):
-                print(f'df_counts of {a_sample}_{sgRNA_name}_{the_signal_formula.__name__}_{operator} exists, loading.\n')
-                df_counts_temp = pd.read_csv(dir_df_counts, index_col=0)
-            else:
-                print(f'For {a_sample}')
-                df_counts_temp, df_noise_temp = load_count(a_sample, regions, bkgs, sgRNA_name = sgRNA_name, signal_formula = the_signal_formula, noise_length=noise_length,
-                                                    ratio_thresh=ratio_thresh, exp_weight=exp_weight, shape_formula=the_shape_formula, operator = operator,
-                                                    region_op = region_op, dirname = temp_dir, pseudo_count=1)
-                print('\n')
-                df_counts_temp['noise_1kb'] = df_noise_temp['noise_bp']*1000
-                df_counts_temp.to_csv(dir_df_counts)
-            dict_df_counts[a_sample] = df_counts_temp
-        # 简化 df_counts 再合并
-        dict_df_counts_sub = {}
-        for a_sample in selected_samples_name:
-            df_counts_sub = dict_df_counts[a_sample].copy()
-            df_counts_sub = df_counts_sub[['chr','st','ed','midpoint','location','ID_1','ID_2','left_signal','right_signal','signal_FC', 'signal_min','shape_bonus','score']].copy()
-            df_counts_sub['sample'] = a_sample
-            dict_df_counts_sub[a_sample] = df_counts_sub
-
-        # 合成 pivot 矩阵
-        df_pivot = pd.concat(dict_df_counts_sub).pivot(index='location', columns='sample', values='score').fillna(0)
-        df_shape = pd.concat(dict_df_counts_sub).pivot(index='location', columns='sample', values='shape_bonus').fillna(0)
-        df_signal_FC = pd.concat(dict_df_counts_sub).pivot(index='location', columns='sample', values='signal_FC').fillna(0)
-        df_signal_min = pd.concat(dict_df_counts_sub).pivot(index='location', columns='sample', values='signal_min').fillna(0)
-
-        df_pivot['std_all'] = df_pivot.std(axis=1)
-        if control_mark != 'none':
-            df_pivot['control_mean'] = df_pivot[ctr_samples].mean(axis=1)
-            df_pivot['control_max']  = df_pivot[ctr_samples].max(axis=1)
-            df_pivot['control_std']  = df_pivot[ctr_samples].std(axis=1)
-            df_pivot['control_signal_FC'] = df_signal_FC[ctr_samples].mean(axis=1)
-            df_pivot['control_signal_min'] = df_signal_min[ctr_samples].mean(axis=1)
-            if repOC == 'mean':
-                df_pivot['control_raw'] = df_pivot['control_mean']
-                df_pivot['control_shape'] = df_shape[ctr_samples].mean(axis=1)
-            elif repOC == 'max':
-                df_pivot['control_raw'] = df_pivot['control_max']
-                df_pivot['control_shape'] = df_shape[ctr_samples].max(axis=1)
-            df_pivot['control_signal'] = df_pivot['control_raw']*df_pivot['control_shape']
-        else:
-            df_pivot['control_signal'] = 'no_control'
-
-        # group mean/min/max
-        if repO == 'mean':
-            df_pivot['exp_raw'] = df_pivot[exp_samples].mean(axis=1)
-            df_pivot['exp_shape'] = df_shape[exp_samples].mean(axis=1)
-        elif repO == 'min':
-            df_pivot['exp_raw'] = df_pivot[exp_samples].min(axis=1)
-            df_pivot['exp_shape'] = df_shape[exp_samples].min(axis=1)
-        elif repO == 'max':
-            df_pivot['exp_raw'] = df_pivot[exp_samples].max(axis=1)
-            df_pivot['exp_shape'] = df_shape[exp_samples].max(axis=1)
-        df_pivot['exp_signal_FC'] = df_signal_FC[exp_samples].mean(axis=1)
-        df_pivot['exp_signal_min'] = df_signal_min[exp_samples].mean(axis=1)
-        df_pivot['exp_signal'] = df_pivot['exp_raw']*df_pivot['exp_shape']
-
-        # 添加坐标 和 ID
-        bed_pivot = bedfmt(df_pivot.index)
-        bed_pivot.index = df_pivot.index
-        df_pivot = pd.concat([bed_pivot,df_pivot], axis=1)
-        df_pivot['midpoint'] = df_pivot['st']+max_region
-
-        point_head = (df_pivot['midpoint']/1000).astype(int)
-        df_pivot['ID_1'] = df_pivot['chr'] + ':' + point_head.astype(str)
-        point_tail = df_pivot['midpoint'] % 1000
-        df_pivot.loc[point_tail<500,'ID_2'] = df_pivot['chr'] + ':' + (point_head-1).astype(str)
-        df_pivot.loc[point_tail>=500,'ID_2'] = df_pivot['chr'] + ':' + (point_head+1).astype(str)
-
-        df_alignment_sub = df_alignment.loc[df_pivot.index, ['best_strand','best_target','target_location','deletion','insertion','mismatch','GG','alignment_score','best_seq_score']]
-        df_pivot = pd.concat([df_pivot,df_alignment_sub],axis=1)
-        df_pivot.to_csv(output)
-    ###########
-    ## score ##
-    ###########
-    
-    output = f'./temp/df_pivot_score_{outname}.csv'
-    if (os.path.isfile(output))&(not args.overwrite):
-        print(f'skip {output}')
-        df_pivot =  pd.read_csv(output, index_col=0)
-    else:
-        df_pivot = pd.read_csv(f'./temp/df_pivot_{outname}.csv', index_col=0)
-        df_pivot[f'raw_score'] = df_pivot[['exp_signal','control_signal']].apply(lambda x: the_score_formula(x['exp_signal'], x['control_signal'], score_pseudo), axis=1 )
-        mean_seq_score = round(df_pivot['best_seq_score'].mean(),3)
-        df_pivot['norm_best_seq_score'] = np.power(df_pivot['best_seq_score']/mean_seq_score, seq_score_power)
-        df_pivot[f'final_score'] = df_pivot[f'raw_score']*df_pivot['norm_best_seq_score']
-        # record the final score before rescue
-        df_pivot['final_score_before_rescue'] = df_pivot['final_score'].copy()
-        
-        # raw rescue
-        df_raw_rescue =  df_pivot[df_pivot['raw_score']>1.5].sort_values(by='raw_score', ascending=False)
-        df_raw_rescue['min_raw'] = df_raw_rescue[exp_samples].min(axis=1)/df_raw_rescue['control_max'].clip(lower=0.5)
-        # 对 min_raw rescue
-        index_raw_rescue = df_raw_rescue[df_raw_rescue['min_raw']>args.RescueThresh1].index
-        # 直接对 raw 进行 rescue
-        index_raw_rescue_2 = df_raw_rescue[df_raw_rescue['raw_score']>args.RescueThresh2].index
-        index_raw_rescue = np.union1d(index_raw_rescue,index_raw_rescue_2)
-        print('rescue',len(index_raw_rescue),'sites')
-        df_pivot['rescue_factor'] = 1
-        df_pivot.loc[index_raw_rescue,'rescue_factor'] = rescue_factor
-        df_pivot.loc[index_raw_rescue,f'final_score'] = df_pivot.loc[index_raw_rescue,f'final_score']*rescue_factor
-        
-        # dedup
-        df_pivot = df_pivot.sort_values(by=f'final_score', ascending=False)
-        list_nondup = dedup_two(df_pivot,'ID_1','ID_2')
-        df_pivot = df_pivot[list_nondup]
-        
-        target_std=0.15
-        df_pivot = df_pivot[df_pivot[f'final_score']>0].copy()
-        n_bkg_sites = int(len(df_pivot)*0.99)
-        score_bkg = df_pivot['final_score'][-n_bkg_sites:]
-        mean_score_bkg = score_bkg.mean()
-        std_score_bkg = score_bkg.std()
-        df_pivot['norm_final_score'] = (df_pivot[f'final_score'] - mean_score_bkg) / std_score_bkg
-        df_pivot['norm_final_score'] = df_pivot[f'norm_final_score']*target_std + 1
-        df_pivot['norm_final_score'] = df_pivot['norm_final_score'].clip(lower=0)
-        df_pivot['log2_norm_final_score'] = np.log2(df_pivot[f'norm_final_score']+1)
-        
-        # fitting normal distribution
-        score_for_fitting = df_pivot['log2_norm_final_score'][-n_bkg_sites:]
-        mu, std = norm.fit(score_for_fitting) 
-        print('mean_score:{:.3f};std:{:.3f}'.format(mu,std))
-        # pv and fdr
-        df_pivot['pv'] = df_pivot[f'log2_norm_final_score'].apply( lambda x: norm.sf(x,loc=mu,scale=std) )
-        df_pivot['pv'].clip(lower=1e-320,inplace=True)
-        df_pivot.to_csv(output)
-    
-    ############
-    ## filter ##
-    ############
-    
-    signal_min_thresh = -0.1
-    signal_min_thresh_2 = -2
-    search_distance = 40000
-    seq_score_thresh = len_sgRNA_PAM*2 - 5*2
-    
-    df_result = df_pivot.copy()
-    ## 
-    candidate_dup = df_result[ df_result[f'exp_signal_min']<=signal_min_thresh ].index[:500]
-    list_dedup = []
-    list_seq_score = []
-    for a_loc in candidate_dup:
-        temp_chr = df_result.loc[a_loc,'chr']
-        temp_seq_score = df_result.loc[a_loc,'best_seq_score']
-        # exp_signal_min 太低直接跳过得分判断
-        if df_result.loc[a_loc,'exp_signal_min'] <= signal_min_thresh_2 :
-            pass
-        else:
-            # 高于 X 分则跳过
-            if temp_seq_score > seq_score_thresh:
-                continue
-        # 取排其前且同chr者
-        temp_df_result = df_result.loc[:a_loc]
-        temp_df_result = temp_df_result[temp_df_result['chr'] == temp_chr].iloc[:-1]
-        if len(temp_df_result)==0:
-            continue
-        else:
-            # 距离小于 search_distance 者记为信号溢出假阳性
-            if (temp_df_result['midpoint']-df_result.loc[a_loc,'midpoint']).abs().min()<search_distance :
-                list_dedup.append(a_loc)
-                list_seq_score.append(temp_seq_score)
-    
-    df_result = df_result[~df_result.index.isin(list_dedup)].copy()
-    print(f'filter {len(list_dedup)} sites')
-    
-    df_result['rank'] = range(1,len(df_result)+1)
-    df_result['fdr'] = fdr(df_result['pv'])
-    df_result.to_csv(f'./temp/df_result_{outname}.csv')
-    
-    df_output = df_result[df_result['log2_norm_final_score']>=1.3].copy()
-    df_output = df_output[['target_location', 'best_strand','best_target','deletion','insertion','mismatch','norm_best_seq_score','norm_final_score', 'log2_norm_final_score','fdr','rank']]
-    df_output.columns = ['target_location', 'strand', 'target', 'deletion', 'insertion', 'mismatch', 'seq_score', 'track_score', 'log2(track_score+1)','FDR', 'rank']
-    df_output.to_csv(f'Trackseq_result_{outname}.csv', index=False)
-    
-    if args.clean:
-        shutil.rmtree('./temp')
-
-    return 'Done!'
-    
-
-if __name__ == '__main__' :
-    result = main()
-    print(result)
-
-