changeset 4:6f0be85be8fb draft

Uploaded
author charles-bernard
date Thu, 27 Oct 2016 06:49:58 -0400
parents 4ada30780811
children 6cfa38b014c1
files data_manager_build_alfa_indexes/data_manager/ALFA.py data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.py data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.xml data_manager_build_alfa_indexes/tool-data/alfa_indexes.loc.sample~ data_manager_build_alfa_indexes/tool_dependencies.xml
diffstat 5 files changed, 1169 insertions(+), 8 deletions(-) [+]
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/data_manager_build_alfa_indexes/data_manager/ALFA.py	Thu Oct 27 06:49:58 2016 -0400
@@ -0,0 +1,1160 @@
+#!/usr/bin/python
+#-*- coding: utf-8 -*-
+
+__author__ = 'noel & bahin'
+''' <decription> '''
+
+import argparse
+import os
+import numpy
+import sys
+import subprocess
+import matplotlib.pyplot as plt
+import matplotlib.cm as cmx
+import matplotlib.colors as colors
+import matplotlib.patheffects as PathEffects
+import re
+from matplotlib.backends.backend_pdf import PdfPages
+# To correctly embbed the texts when saving plots in svg format
+import matplotlib
+matplotlib.rcParams['svg.fonttype'] = 'none'
+
+##########################################################################
+#                         FUNCTIONS                                      #
+##########################################################################
+
+def init_dict(d, key, init):
+	if key not in d:
+		d[key] = init
+
+def get_chromosome_lengths(args):
+	"""
+	Parse the file containing the chromosomes lengths.
+	If no length file is provided, browse the annotation file (GTF) to estimate the chromosome sizes (
+	"""
+	lengths={}
+	gtf_chrom_names=set()
+	force_get_lengths = False
+	# If the user provided the chromosome length file
+	if args.chr_len:
+		with open(args.chr_len, 'r') as chr_len_file:
+			for line in chr_len_file:
+				lengths[line.split('\t')[0]] = int(line.rstrip().split('\t')[1])
+		with open(args.annotation,'r') as gtf_file:
+			for line in gtf_file:
+				if not line.startswith('#'):
+					chrom = line.split('\t')[0]
+					if chrom not in gtf_chrom_names:
+						gtf_chrom_names.add(chrom)
+		for chrom in lengths.keys():
+			if chrom not in gtf_chrom_names:
+				print "Warning: at least one chromosome name ('"+chrom+"') of the file '"+args.chr_len+"'does not match any chromosome name if GTF and was ignored."
+				#del lengths[chrom]
+				break
+		for chrom in gtf_chrom_names:
+			if force_get_lengths: break
+			if chrom not in lengths.keys():
+				print "WARNING: chromosome name '"+chrom+"' was found in gtf and does not match any chromosome name provided in",args.chr_len+". "
+				print "\t=> The chromosome lenghts will be approximated using annotations in the GTF file."
+				continue_value =""
+				while continue_value not in {"yes","y","no","n"}:
+					continue_value = raw_input("\tDo you want to continue ('yes' or 'y')?\n\tElse write 'no' or 'n' to exit the script and check your file of lengths.\n")
+					if continue_value == "no" or continue_value == "n":
+						sys.exit("Exiting")
+					elif continue_value == "yes" or continue_value == "y":
+						force_get_lengths = True
+						break
+					print "Error: use 'yes/y/no/n' only"
+		if not force_get_lengths:
+			return lengths
+	# Otherwise, (or if at least one chromosome was missing in chromosome lengths file) we consider the end of the last annotation of the chromosome in the GTF file as the chromosome length
+	with open(args.annotation, 'r') as gtf_file:
+		for line in gtf_file:
+			if not line.startswith('#'):
+				chrom = line.split('\t')[0]
+				end = int(line.split('\t')[4])
+				init_dict(lengths, chrom, 0)
+				lengths[chrom] = max(lengths[chrom], end)
+		if force_get_lengths:
+			print "The chromosome lenghts have been approximated using the last annotations in the GTF file."
+		return lengths
+
+def write_feature_on_index(feat,chrom, start, stop, sign, stranded_genome_index, unstranded_genome_index=None):
+	grouped_by_biotype_features = []
+	for biotype,categs in feat.iteritems():
+		categ_list=[]
+		for cat in set(categs):
+			categ_list.append(cat)
+		grouped_by_biotype_features.append(":".join((str(biotype),",".join(categ_list))))
+	stranded_genome_index.write('\t'.join((chrom, start, stop, sign,''))+'\t'.join(grouped_by_biotype_features)+'\n')
+	if unstranded_genome_index :
+		unstranded_genome_index.write('\t'.join((chrom, start, stop, '.',''))+'\t'.join(grouped_by_biotype_features)+'\n')
+
+
+def add_info(cpt, feat_values, start, stop, chrom=None, unstranded_genome_index=None, stranded_genome_index = None , biotype_prios=None, coverage=1, categ_prios=None):
+	"""
+	From an annotated genomic interval (start/stop positions and associated feature : one or more category(ies)/biotype pair(s) )
+	- If a file is provided: write the  information at the end of the index file being generated;
+	- else : browse the features and update the counts of categories/biotypes found in the genome.
+	"""
+	## Writing in the file is it was provided
+	if stranded_genome_index :
+		unstranded_features = None
+		# If only one strand has a feature, this feature will directly be written on the unstranded index
+		if feat_values[0] == {} :
+			# A interval with no feature corresponds to a region annotated only on the reverse strand : update 'antisense' counts
+			stranded_genome_index.write('\t'.join((chrom, start, stop, '+','antisense\n')))
+			write_feature_on_index(feat_values[1], chrom ,start, stop, '-', stranded_genome_index, unstranded_genome_index)
+		else :
+			if feat_values[1] == {} :
+				write_feature_on_index(feat_values[0], chrom ,start, stop, '+', stranded_genome_index, unstranded_genome_index)
+				stranded_genome_index.write('\t'.join((chrom, start, stop, '-','antisense\n')))
+			
+		# Else, the unstranded index should contain the union of plus and minus features
+			else :
+				write_feature_on_index(feat_values[0], chrom ,start, stop, '+', stranded_genome_index)
+				write_feature_on_index(feat_values[1], chrom ,start, stop, '-', stranded_genome_index)
+				unstranded_feat = dict(feat_values[0], **feat_values[1])
+				for name in set(feat_values[0]) & set(feat_values[1]):
+					unstranded_feat[name]+=feat_values[0][name]
+				write_feature_on_index(unstranded_feat, chrom ,start, stop, '.', unstranded_genome_index)
+	
+	## Increasing category counter(s)
+	else :
+		# Default behavior if no biotype priorities : category with the highest priority for each found biotype has the same weight (1/n_biotypes)
+		if not biotype_prios:
+			nb_feat = len(feat_values)
+			# For every categ(s)/biotype pairs
+			for feat in feat_values:
+				cur_prio = 0
+				selected_categ = []
+				# Separate categorie(s) and biotype
+				try:
+					biot,cats = feat.split(":")
+				# Error if feature equal "antisense" : update the 'antisense/antisense' counts
+				except ValueError :
+					try :
+						cpt[(feat,feat)] += (int(stop) - int(start)) * coverage / float(nb_feat) 
+					except :
+						cpt[(feat,feat)] = (int(stop) - int(start)) * coverage / float(nb_feat)
+					return
+				# Browse the categories and get only the one(s) with highest priority
+				for cat in cats.split(','):
+					try: prio = prios[cat]
+					except:
+						#TODO Find a way to add unknown categories
+						if cat not in unknown_feature:
+							print >> sys.stderr, "Warning: Unknown categorie %s found and ignored\n...\r" %cat,
+							unknown_feature.append(cat)
+						continue
+					if prio > cur_prio :
+						cur_prio = prio
+						selected_categ = [cat]
+					if prio == cur_prio :
+						if cat != selected_categ :
+							try:
+								if cat not in selected_categ :
+									selected_categ.append(cat)
+							except TypeError :
+								selected_categ = [selected_categ,cat]
+				# Increase each counts by the coverage divided by the number of categories and biotypes
+				nb_cats = len(selected_categ)
+				for cat in selected_categ :
+					try:
+						cpt[(cat,biot)] += (int(stop) - int(start)) * coverage / (float(nb_feat * nb_cats))
+					except KeyError:
+						cpt[(cat,biot)] = (int(stop) - int(start)) * coverage / (float(nb_feat * nb_cats))
+					#else :
+						#cpt[(cats,biot)] = (int(stop) - int(start)) / float(nb_feat) * coverage
+		# Else, apply biotype selection according to the priority set
+		else:
+			#TODO Add an option to pass biotype priorities and handle it
+			pass
+
+def print_chrom(features_dict, chrom, stranded_index_file, unstranded_index_file, cpt_genome):
+	with open(unstranded_index_file,'a') as findex, open(stranded_index_file,'a') as fstrandedindex:
+		# Initialization of variables : start position of the first interval and associated features for +/- strands
+		start = ""
+		for pos in sorted(features_dict['+'].keys()):
+			if start != "":
+				add_info(cpt_genome, [features_plus,features_minus], str(start), str(pos), chrom, stranded_genome_index = fstrandedindex, unstranded_genome_index = findex)
+			start = pos
+			features_plus = features_dict['+'][pos]
+			features_minus = features_dict['-'][pos]
+
+
+def create_genome_index(annotation, unstranded_genome_index, stranded_genome_index,cpt_genome,prios,biotypes,chrom_sizes):
+	''' Create an index of the genome annotations and save it in a file'''
+	print '\n### Generating genome indexes\n',
+	sys.stdout.flush()
+	# Initializations
+	intervals_dict = 0
+	max_value = -1
+	prev_chrom = ''
+	reverse_strand = {'+':'-','-':'+'}
+	i = 0 # Line counter
+	# Write the chromosomes lengths as comment lines before the genome index
+	with open(unstranded_genome_index,'w') as findex, open(stranded_genome_index,'w') as fstrandedindex:
+		for key,value in chrom_sizes.items():
+			findex.write("#%s\t%s\n" %(key, value))
+			fstrandedindex.write("#%s\t%s\n" %(key, value))
+	# Running through the GTF file and writing into genome index files
+	with open(annotation, 'r') as gtf_file:
+		for line in gtf_file:
+			# Print messages after X processed lines
+			i += 1
+			if i % 100000 == 0:
+				print >> sys.stderr, '\r%s line processed...' %str(i)
+				print '\r                          \r. . .',
+				sys.stdout.flush()
+			elif i % 20000 == 0:
+				print '\r                          \r. . .',
+				sys.stdout.flush()
+			elif i % 2000 == 0:
+				print '.',
+				sys.stdout.flush()
+			# Processing lines except comment ones
+			if not line.startswith('#'):
+				# Getting the line infos
+				line_split=line[:-1].split('\t')
+				chrom = line_split[0]
+				cat=line_split[2]
+				start = int(line_split[3]) - 1
+				stop = int(line_split[4])
+				strand = line_split[6]
+				antisense = reverse_strand[strand]
+				biotype=line_split[8].split('biotype')[1].split(';')[0].strip('" ')
+				feat = [(cat,biotype)]
+				# Registering chromosome info in the genome index file if this is a new chromosome or a annotation not overlapping previously recorded features
+				if start > max_value or chrom != prev_chrom: 
+					# Write the previous features
+					if intervals_dict != 0:
+						if chrom != prev_chrom :
+							print_chrom(intervals_dict, prev_chrom, stranded_genome_index, unstranded_genome_index, cpt_genome)
+							print "\rChromosome '" + prev_chrom + "' registered."
+						else:
+							print_chrom(intervals_dict, chrom, stranded_genome_index, unstranded_genome_index, cpt_genome)
+					prev_chrom = chrom
+					# (Re)Initializing the chromosome lists
+					intervals_dict = {strand:{start:{biotype:[cat]}, stop:{}},antisense:{start:{},stop:{}}}
+					max_value = stop
+					
+				# Update the dictionary which represents intervals for every disctinct annotation 
+				else:
+					# Get intervals on the same strand as the current feature
+					stranded_intervals = intervals_dict[strand]
+					start_added = False
+					for pos in sorted(stranded_intervals.keys()):
+						#print pos
+						#print stranded_intervals[pos]
+						#print
+						# While the position is below the feature's interval, store the features
+						if pos < start :
+							cur_cat_dict = dict(stranded_intervals[pos])
+							cur_antisense_dict = dict(intervals_dict[antisense][pos])
+							
+						# If the start position already exists: update it by addind the new feature
+						elif pos == start :
+							cur_cat_dict = dict(stranded_intervals[pos])
+							cur_antisense_dict = dict(intervals_dict[antisense][pos])
+							#print "cur",cur_cat_dict
+							try : stranded_intervals[pos][biotype] = stranded_intervals[pos][biotype]+[cat]
+							except KeyError: stranded_intervals[pos][biotype] = [cat]
+							start_added = True
+							#print "cur",cur_cat_dict
+							
+						elif pos > start :
+							# Create a new entry for the start position if necessary
+							if not start_added :
+								#print "cur",cur_cat_dict
+								stranded_intervals[start] = dict(cur_cat_dict)
+								stranded_intervals[start][biotype] = [cat]
+								#stranded_intervals[pos][biotype].append(cat)
+								intervals_dict[antisense][start] = cur_antisense_dict
+								start_added = True
+								#print "cur",cur_cat_dict
+							# While the position is below the stop, just add the new feature
+							if pos < stop :
+								cur_cat_dict = dict(stranded_intervals[pos])
+								cur_antisense_dict = dict(intervals_dict[antisense][pos])
+								try: stranded_intervals[pos][biotype] = stranded_intervals[pos][biotype] + [cat]
+								except KeyError: stranded_intervals[pos][biotype] = [cat]
+							# Close the created interval : create an entry at the stop position and restore the features
+							elif pos > stop :
+								stranded_intervals[stop] = dict(cur_cat_dict)
+								intervals_dict[antisense][stop] = cur_antisense_dict
+								break
+							else :
+								break
+						#try:
+							#cur_cat_dict = list(stranded_intervals[pos][biotype])
+						#except KeyError: cur_cat_dict = list()
+						#print stranded_intervals[pos]
+						#print
+					# Extend the dictionary if necessary
+					if stop > max_value:
+						max_value = stop
+						stranded_intervals[stop] = {}
+						intervals_dict[antisense][stop] = {}
+					#except KeyError:
+						#print intervals_dict
+						#quit()
+						#intervals_dict[strand] = {strand:{start:{biotype:[cat]}, stop:{}},antisense:{start:{},stop:{}}}
+						#continue
+					#for sign in ['-','+']:
+						#print sign
+						#for key,val in sorted(intervals_dict[sign].iteritems()):
+							#print key,'\t',val
+						#print
+					#print "-------\n"
+
+								
+		#Store the categories of the last chromosome
+		print_chrom(intervals_dict, chrom, stranded_genome_index, unstranded_genome_index, cpt_genome)
+		print "\rChromosome '" + prev_chrom + "' registered.\nDone!"
+
+
+def create_bedgraph_files(bams,strandness):
+	samples_files = []
+	labels = []
+	print "\n### Generating the bedgraph files"
+	for n in range(0, len(bams), 2):
+		print "\rProcessing '%s'\n..." %bams[n],
+		sys.stdout.flush()
+		#Get the label for this sample
+		label = bams[n+1]
+		#Modify it to contain only alphanumeric caracters (avoid files generation with dangerous names)
+		modified_label = "_".join(re.findall(r"[\w']+", label))
+		if strandness in ["reverse","fr-secondstrand"]:
+			subprocess.call('bedtools genomecov -bg -split -strand - -ibam ' + bams[n] + ' > ' + modified_label + '.plus.bedgraph', shell=True)
+			subprocess.call('bedtools genomecov -bg -split -strand + -ibam ' + bams[n] + ' > ' + modified_label + '.minus.bedgraph', shell=True)
+		elif strandness in ["forward","fr-firststrand"]:
+			subprocess.call('bedtools genomecov -bg -split -strand + -ibam ' + bams[n] + ' > ' + modified_label + '.plus.bedgraph', shell=True)
+			subprocess.call('bedtools genomecov -bg -split -strand - -ibam ' + bams[n] + ' > ' + modified_label + '.minus.bedgraph', shell=True)
+		else :
+			subprocess.call('bedtools genomecov -bg -split -ibam ' + bams[n] + ' > ' + modified_label + '.bedgraph', shell=True)
+		samples_files.append(modified_label)
+		labels.append(label)
+	print "\rDone!"
+	return samples_files, labels
+
+def read_gtf(gtf_index_file, sign):
+	global gtf_line, gtf_chrom, gtf_start, gtf_stop, gtf_features, endGTF
+	strand = ""
+	while strand != sign :
+		gtf_line = gtf_index_file.readline()
+		# If the GTF file is finished
+		if not gtf_line:
+			endGTF = True
+			return endGTF
+		splitline = gtf_line.rstrip().split('\t')
+		try: strand = splitline[3]
+		# strand information can not be found in the file file header
+		except IndexError: pass
+	gtf_chrom = splitline[0]
+	gtf_features = splitline[4:]
+	gtf_start, gtf_stop = map(int, splitline[1:3])
+	return endGTF
+
+def read_counts_files(counts_files):
+	cpt={}
+	cpt_genome={}
+	labels=[]
+	for fcounts in counts_files:
+		label=os.path.splitext(os.path.basename(fcounts))[0]
+		labels.append(label)
+		cpt[label]={}
+		with open (fcounts,"r") as f:
+			for line in f:
+				if line[0]=="#":
+					continue
+				line_split=line[:-1].split('\t')
+				feature=tuple(line_split[0].split(','))
+				cpt[label][feature]=float(line_split[1])
+				cpt_genome[feature]=float(line_split[2])
+	return cpt,cpt_genome,labels
+
+
+def get_chromosome_names_in_index(genome_index):
+		chrom_list = []
+		with open(genome_index, 'r') as findex:
+			chrom = ""
+			for line in findex:
+				cur_chrom = line.split('\t')[0]
+				if cur_chrom == chrom:
+					pass
+				else:
+					chrom = cur_chrom
+					if chrom not in chrom_list:
+						chrom_list.append(chrom)
+		return set(chrom_list)
+
+
+def intersect_bedgraphs_and_index_to_counts_categories(samples_files,samples_names,prios,genome_index, strandness, biotype_prios = None):
+	global gtf_line, gtf_chrom, gtf_start, gtf_stop, gtf_cat, endGTF
+	print "\n### Intersecting files with indexes"
+	unknown_chrom = []
+	cpt = {} # Counter for the nucleotides in the BAM input file(s)
+	for n in range(len(samples_files)):
+		sample_file=samples_files[n]
+		sample_name=samples_names[n]
+		# Initializing the category counter dict for this sample
+		init_dict(cpt, sample_name, {})
+		if strandness == "unstranded":
+			strands = [("",".")]
+		else:
+			strands = [('.plus','+'), ('.minus','-')]
+			
+		# Intersecting the BEDGRAPH and genome index files
+		print "\rProcessing '%s'\n. . ." %sample_file,
+		sys.stdout.flush()
+
+		for strand,sign in strands:
+			prev_chrom = ''
+			endGTF = False # Reaching the next chr or the end of the GTF index
+			intergenic_adds = 0.0
+			i = 0
+			i_chgt = 0
+			with open(sample_file + strand + '.bedgraph', 'r') as bam_count_file:
+				# Running through the BEDGRAPH file
+				for bam_line in bam_count_file:
+					i += 1
+					if i % 10000 == 0:
+						print ".",
+						sys.stdout.flush()
+					if i % 100000 == 0:
+						print "\r                              \r. . .",
+						sys.stdout.flush()
+					# Getting the BAM line info
+					bam_chrom = bam_line.split('\t')[0]
+					bam_start, bam_stop, bam_cpt = map(float, bam_line.split('\t')[1:4])
+					# Skip the line if the chromosome is not in the index
+					if bam_chrom not in chrom_list:
+						if bam_chrom not in unknown_chrom:
+							unknown_chrom.append(bam_chrom)
+							print "\r                          \r Chromosome '" + bam_chrom + "' not found in index."
+						continue
+					# If this is a new chromosome (or the first one)
+					if bam_chrom != prev_chrom:
+						i_chgt = i
+						intergenic_adds = 0.0
+						# (Re)opening the GTF index and looking for the first line of the matching chr
+						try: gtf_index_file.close()
+						except UnboundLocalError: pass
+						gtf_index_file = open(genome_index, 'r')
+						endGTF = False
+						read_gtf(gtf_index_file, sign)
+						while bam_chrom != gtf_chrom:
+							read_gtf(gtf_index_file, sign)
+							if endGTF:
+								break
+						prev_chrom = bam_chrom
+
+					# Looking for the first matching annotation in the GTF index
+					while (not endGTF) and (gtf_chrom == bam_chrom) and (bam_start >= gtf_stop):
+						read_gtf(gtf_index_file, sign)
+						if gtf_chrom != bam_chrom:
+							endGTF = True
+					# Processing BAM lines before the first GTF annotation if there are
+					if bam_start < gtf_start:
+						# Increase the 'intergenic' category counter with all or part of the BAM interval
+						try:
+							intergenic_adds += min(bam_stop,gtf_start)-bam_start
+							cpt[sample_name][('intergenic','intergenic')] += (min(bam_stop, gtf_start) - bam_start) * bam_cpt
+						except KeyError:
+							cpt[sample_name][('intergenic','intergenic')] = (min(bam_stop, gtf_start) - bam_start) * bam_cpt
+						# Go to next line if the BAM interval was totally upstream the first GTF annotation, carry on with the remaining part otherwise
+						if endGTF or (bam_stop <= gtf_start):
+							continue
+						else:
+							bam_start = gtf_start
+
+					# We can start the crossover
+					while not endGTF:
+						# Update category counter
+						add_info(cpt[sample_name], gtf_features, bam_start, min(bam_stop,gtf_stop), coverage = bam_cpt, categ_prios = prios)
+						# Read the next GTF file line if the BAM line is not entirely covered
+						if bam_stop > gtf_stop:
+							# Update the BAM start pointer
+							bam_start = gtf_stop
+							endGTF = read_gtf(gtf_index_file, sign)
+							# If we read a new chromosome in the GTF file then it is considered finished
+							if bam_chrom != gtf_chrom:
+								endGTF = True
+							if endGTF:
+								break
+						else:
+							# Next if the BAM line is entirely covered
+							bam_start = bam_stop
+							break
+
+					# Processing the end of the BAM line if necessary
+					if endGTF and (bam_stop > bam_start):
+						try:
+							cpt[sample_name][('intergenic','intergenic')] += (bam_stop - bam_start) * bam_cpt
+						except KeyError:
+							cpt[sample_name][('intergenic','intergenic')] = (bam_stop - bam_start) * bam_cpt
+				gtf_index_file.close()
+	print "\r                             \rDone!"
+	return cpt
+
+def write_counts_in_files(cpt,genome_counts):
+	for label,dico in cpt.items():
+		label = "_".join(re.findall(r"[\w']+", label))
+		with open(label+".categories_counts","w") as fout:
+			fout.write("#Category,biotype\tCounts_in_bam\tSize_in_genome\n" )
+			for feature,counts in dico.items():
+				fout.write("%s\t%s\t%s\n" %(','.join(feature),counts,genome_counts[feature]))
+
+def recategorize_the_counts(cpt,cpt_genome,final):
+	final_cat_cpt={}
+	final_genome_cpt={}
+	for f in cpt:
+		#print "\nFinal categories for",f,"sample"
+		final_cat_cpt[f]={}
+		for final_cat in final:
+			tot = 0
+			tot_genome=0
+			for cat in final[final_cat]:
+					tot += cpt[f][cat]
+					tot_genome+=cpt_genome[cat]
+			#output_file.write('\t'.join((final_cat, str(tot))) + '\n')
+			#print '\t'.join((final_cat, str(tot)))
+			final_cat_cpt[f][final_cat]=tot
+			final_genome_cpt[final_cat]=tot_genome
+	return final_cat_cpt,final_genome_cpt
+
+def group_counts_by_categ(cpt,cpt_genome,final,selected_biotype):
+	final_cat_cpt={}
+	final_genome_cpt={}
+	filtered_cat_cpt = {}
+	for f in cpt:
+		final_cat_cpt[f]={}
+		filtered_cat_cpt[f] = {}
+		for final_cat in final:
+			tot = 0
+			tot_filter = 0
+			tot_genome=0
+			for cat in final[final_cat]:
+					for key,value in cpt[f].items():
+						if key[0] == cat :
+							tot += value
+							tot_genome+=cpt_genome[key]
+							if key[1] == selected_biotype :
+								tot_filter += value
+			#output_file.write('\t'.join((final_cat, str(tot))) + '\n')
+			#print '\t'.join((final_cat, str(tot)))
+			final_cat_cpt[f][final_cat]=tot
+			if tot_genome == 0:
+				final_genome_cpt[final_cat]= 1e-100
+			else:
+				final_genome_cpt[final_cat]=tot_genome
+			filtered_cat_cpt[f][final_cat]=tot_filter
+	#if 'antisense' in final_genome_cpt: final_genome_cpt['antisense'] = 0
+	return final_cat_cpt,final_genome_cpt,filtered_cat_cpt
+
+def group_counts_by_biotype(cpt,cpt_genome,biotypes):
+	final_cpt={}
+	final_genome_cpt={}
+	for f in cpt:
+		final_cpt[f]={}
+		for biot in biotypes:
+			tot = 0
+			tot_genome=0
+			try:
+				for final_biot in biotypes[biot]:
+					for key,value in cpt[f].items():
+						if key[1] == final_biot :
+							tot += value
+							#if key[1] != 'antisense':
+							tot_genome+=cpt_genome[key]
+			except:
+				for key,value in cpt[f].items():
+					if key[1] == biot :
+						tot += value
+						tot_genome+=cpt_genome[key]
+			if tot != 0:
+				final_cpt[f][biot]=tot
+				final_genome_cpt[biot]=tot_genome
+	return final_cpt,final_genome_cpt
+
+#def get_cmap(N):
+	#'''Returns a function that maps each index in 0, 1, ... N-1 to a distinct
+	#RGB color.'''
+	#color_norm  = colors.Normalize(vmin=0, vmax=N-1)
+	#scalar_map = cmx.ScalarMappable(norm=color_norm, cmap='hsv')
+	#def map_index_to_rgb_color(index):
+		#return scalar_map.to_rgba(index)
+	#return map_index_to_rgb_color
+
+def one_sample_plot(ordered_categs, percentages, enrichment, n_cat, index, index_enrichment, bar_width, counts_type, title) :
+	### Initialization
+	fig = plt.figure(figsize=(13,9))
+	ax1 = plt.subplot2grid((2,4),(0,0),colspan=2)
+	ax2 = plt.subplot2grid((2,4),(1,0),colspan=2)
+	cmap= plt.get_cmap('Spectral')
+	cols=[cmap(x) for x in xrange(0,256,256/n_cat)]
+	if title:
+		ax1.set_title(title+"in: %s" %samples_names[0])
+	else :
+		ax1.set_title(counts_type+" distribution in mapped reads in: %s" %samples_names[0])
+	ax2.set_title('Normalized counts of '+counts_type)
+
+	
+	### Barplots
+	#First barplot: percentage of reads in each categorie
+	ax1.bar(index, percentages, bar_width,
+				color=cols)
+	#Second barplot: enrichment relative to the genome for each categ
+	# (the reads count in a categ is divided by the categ size in the genome)
+	ax2.bar(index_enrichment, enrichment, bar_width,
+				color=cols,)
+	### Piecharts
+	pielabels = [ordered_categs[i] if percentages[i]>0.025 else '' for i in xrange(n_cat)]
+	sum_enrichment = numpy.sum(enrichment)
+	pielabels_enrichment = [ordered_categs[i] if enrichment[i]/sum_enrichment>0.025 else '' for i in xrange(n_cat)]
+	# Categories piechart
+	ax3 = plt.subplot2grid((2,4),(0,2))
+	pie_wedge_collection, texts = ax3.pie(percentages,labels=pielabels, shadow=True, colors=cols)
+	# Enrichment piechart
+	ax4 = plt.subplot2grid((2,4),(1,2))
+	pie_wedge_collection, texts = ax4.pie(enrichment,labels=pielabels_enrichment, shadow=True, colors=cols)
+	# Add legends (append percentages to labels)
+	labels = [" ".join((ordered_categs[i],"({:.1%})".format(percentages[i]))) for i in range(len(ordered_categs))]
+	ax3.legend(pie_wedge_collection,labels,loc='center',fancybox=True, shadow=True,prop={'size':'medium'}, bbox_to_anchor=(1.7,0.5))
+	labels = [" ".join((ordered_categs[i],"({:.1%})".format(enrichment[i]/sum_enrichment))) for i in range(len(ordered_categs))]# if ordered_categs[i] != 'antisense']
+	ax4.legend(pie_wedge_collection,labels,loc='center',fancybox=True, shadow=True,prop={'size':'medium'},bbox_to_anchor=(1.7,0.5))
+	# Set aspect ratio to be equal so that pie is drawn as a circle
+	ax3.set_aspect('equal')
+	ax4.set_aspect('equal')
+	return fig, ax1, ax2
+
+def make_plot(ordered_categs,samples_names,categ_counts,genome_counts,pdf, counts_type, threshold, title = None ,svg = None, png = None):
+	# From ordered_categs, keep only the features (categs or biotypes) that we can find in at least one sample.
+	existing_categs = set()
+	for sample in categ_counts.values():
+		existing_categs |= set(sample.keys())
+	ordered_categs = filter(existing_categs.__contains__, ordered_categs)
+	n_cat = len(ordered_categs)
+	n_exp=len(samples_names)
+	##Initialization of the matrix of counts (nrow=nb_experiements, ncol=nb_categorie)
+	counts=numpy.matrix(numpy.zeros(shape=(n_exp,n_cat)))
+	for exp in xrange(len(samples_names)):
+		for cat in xrange(len(ordered_categs)):
+			try:	counts[exp,cat]=categ_counts[samples_names[exp]][ordered_categs[cat]]
+			except:	pass
+
+	##Normalize the categorie sizes by the total size to get percentages
+	sizes=[]
+	sizes_sum=0
+	for cat in ordered_categs:
+		sizes.append(genome_counts[cat])
+		sizes_sum+=genome_counts[cat]
+	if 'antisense' in ordered_categs:
+		antisense_pos = ordered_categs.index('antisense')
+		sizes[antisense_pos] = 1e-100
+	for cpt in xrange(len(sizes)):
+		sizes[cpt]/=float(sizes_sum)
+
+	## Create array which contains the percentage of reads in each categ for every sample
+	percentages=numpy.array(counts/numpy.sum(counts,axis=1))
+	## Create the enrichment array (counts divided by the categorie sizes in the genome)
+	enrichment=numpy.array(percentages/sizes)
+	if 'antisense_pos' in locals(): 
+		for i in xrange(len(samples_names)):
+			enrichment[i][antisense_pos] = 0
+	#enrichment=numpy.log(numpy.array(percentages/sizes))
+	for exp in xrange(n_exp):
+		for i in xrange(n_cat):
+			val = enrichment[exp][i]
+			if val > 1: 
+				enrichment[exp][i] = val-1
+			elif val == 1 or val == 0:
+				enrichment[exp][i] = 0
+			else:
+				enrichment[exp][i] = -1/val+1
+
+	#### Finally, produce the plot
+
+	##Get the colors from the colormap
+	ncolor=16
+	cmap = ["#e47878", "#68b4e5", "#a3ea9b", "#ea9cf3", "#e5c957", "#a3ecd1", "#e97ca0", "#66d985", "#8e7ae5", "#b3e04b", "#b884e4", "#e4e758", "#738ee3", "#e76688", "#70dddd", "#e49261"]
+	if n_exp > ncolor:
+		cmap = plt.get_cmap('Set3',n_exp)
+		cmap = [cmap(i) for i in xrange(n_exp)]
+	
+	## Parameters for the plot
+	opacity = 1
+	#Create a vector which contains the position of each bar
+	index = numpy.arange(n_cat)
+	#Size of the bars (depends on the categs number)
+	bar_width = 0.9/n_exp
+
+
+	##Initialise the subplot
+	# if there is only one sample, also plot piecharts 
+	#if n_exp == 1 and counts_type.lower() == 'categories':
+		#fig, ax1, ax2 = one_sample_plot(ordered_categs, percentages[0], enrichment[0], n_cat, index, bar_width, counts_type, title)
+	## If more than one sample
+	#else:
+	fig, (ax1,ax2) = plt.subplots(2,figsize=(5+(n_cat+2*n_exp)/3,10))
+	# Store the bars objects for enrichment plot
+	rects = []
+	#For each sample/experiment
+	for i in range(n_exp):
+		#First barplot: percentage of reads in each categorie
+		ax1.bar(index+i*bar_width, percentages[i], bar_width,
+					alpha=opacity,
+					color=cmap[i],
+					label=samples_names[i], edgecolor='#FFFFFF', lw=0)
+		#Second barplot: enrichment relative to the genome for each categ
+		# (the reads count in a categ is divided by the categ size in the genome)
+		rects.append( ax2.bar(index+i*bar_width, enrichment[i], bar_width,
+					alpha=opacity,
+					color=cmap[i],
+					label=samples_names[i], edgecolor=cmap[i], lw=0))
+		
+	## Graphical options for the plot
+	# Adding of the legend
+	ax1.legend(loc='best',frameon=False)
+	#ax2.legend(loc='upper center',bbox_to_anchor=(0.5,-0.1), fancybox=True, shadow=True)
+	# Main titles
+	if title:
+		ax1.set_title(title)
+	else :
+		ax1.set_title(counts_type+" distribution in mapped reads")
+	ax2.set_title('Normalized counts of '+counts_type)
+			
+	# Adding enrichment baseline
+	#ax2.axhline(y=0,color='black',linestyle='dashed',linewidth='1.5')
+	# Axes limits
+	ax1.set_xlim(-0.1,len(ordered_categs)+0.1)
+	if len(sizes) == 1: ax1.set_xlim(-2,3) 
+	ax2.set_xlim(ax1.get_xlim())
+	# Set axis limits (max/min values + 5% margin)
+	ax2_ymin = []
+	ax2_ymax = []
+	for sample_values in enrichment:
+		ax2_ymin.append(min(sample_values))
+		ax2_ymax.append(max(sample_values))
+	ax2_ymax = max(ax2_ymax)
+	ax2_ymin = min(ax2_ymin)
+	margin_top, margin_bottom = (abs(0.05*ax2_ymax), abs(0.05*ax2_ymin))
+	ax1.set_ylim(0,ax1.get_ylim()[1]*1.05)
+	if threshold:
+		threshold_bottom = -abs(float(threshold[0]))+1
+		threshold_top = float(threshold[1])-1
+		
+		for i in xrange(n_exp):
+			for y in xrange(n_cat):
+				val = enrichment[i][y] 
+				if not numpy.isnan(val) and not (threshold_bottom < val < threshold_top):
+					rect = rects[i][y]
+					rect_height = rect.get_height()
+					if rect.get_y() < 0:
+						diff = rect_height + threshold_bottom
+						rect.set_y(threshold_bottom)
+						ax2_ymin = threshold_bottom
+						margin_bottom = 0
+					else:
+						diff = rect_height - threshold_top
+						ax2_ymax = threshold_top
+						margin_top = 0
+					rect.set_height(rect.get_height()-diff)
+	if margin_top != 0 and margin_bottom != 0:
+		margin_top, margin_bottom = [max(margin_top, margin_bottom) for i in xrange(2)]
+	ax2.set_ylim(ax2_ymin-margin_bottom,ax2_ymax+margin_top)
+	# Y axis title
+	ax1.set_ylabel('Proportion of reads (%)')
+	ax2.set_ylabel('Enrichment relative to genome')
+	# X axis title
+	ax1.set_xlabel(counts_type)
+	ax2.set_xlabel(counts_type)
+	# Add the categories on the x-axis
+	ax1.set_xticks(index + bar_width*n_exp/2)
+	ax2.set_xticks(index + bar_width*n_exp/2)
+	if counts_type.lower() != 'categories':
+		ax1.set_xticklabels(ordered_categs,rotation='30',ha='right')
+		ax2.set_xticklabels(ordered_categs,rotation='30',ha='right')
+	else:
+		ax1.set_xticklabels(ordered_categs)
+		ax2.set_xticklabels(ordered_categs)
+	# Display fractions values in percentages
+	ax1.set_yticklabels([str(int(i*100)) for i in ax1.get_yticks()])
+	# Correct y-axis ticks labels for enrichment subplot
+	#ax2.set_yticklabels([str(i+1)+"$^{+1}$" if i>0 else 1 if i==0 else str(-(i-1))+"$^{-1}$" for i in ax2.get_yticks()])
+	yticks = list(ax2.get_yticks())
+	yticks = [ yticks[i]-1 if yticks[i]>9 else yticks[i]+1 if yticks[i]<-9 else yticks[i] for i in xrange(len(yticks))] 
+	ax2.set_yticks(yticks)
+	ax2.set_yticklabels([str(int(i+1))+"$^{+1}$" if i>0 and i%1==0 else str(i+1)+"$^{+1}$" if i>0 else 1 if i==0 else str(int(-(i-1)))+"$^{-1}$" if i<0 and i%1==0 else str(-(i-1))+"$^{-1}$" for i in ax2.get_yticks()])
+	#ax2.set_yticklabels([i+1 if i>0 else 1 if i==0 else "$\\frac{1}{%s}$" %-(i-1) for i in ax2.get_yticks()])
+	# Change appearance of 'antisense' bars on enrichment plot since we cannot calculate an enrichment for this artificial category
+	if 'antisense_pos' in locals(): #ax2.text(antisense_pos+bar_width/2,ax2.get_ylim()[1]/10,'NA')
+		for i in xrange(n_exp) :
+			rect = rects[i][antisense_pos]
+			rect.set_y(ax2.get_ylim()[0])
+			rect.set_height(ax2.get_ylim()[1] - ax2.get_ylim()[0])
+			rect.set_hatch('/')
+			rect.set_fill(False)
+			rect.set_linewidth(0)
+			#rect.set_color('lightgrey')
+			#rect.set_edgecolor('#EDEDED')
+			rect.set_color('#EDEDED')
+		ax2.text(index[antisense_pos] + bar_width*n_exp/2 - 0.1, (ax2_ymax+ax2_ymin)/2, 'NA')
+	# Add text for features absent in sample
+	for i in xrange(n_exp):
+		for y in xrange(n_cat):
+			if percentages[i][y] == 0:
+				txt = ax1.text(y +  bar_width*(i+0.5), 0.02, 'Absent in sample', rotation = 'vertical', color = cmap[i], horizontalalignment ='center', verticalalignment = 'bottom')
+				txt.set_path_effects([PathEffects.Stroke(linewidth=0.5),PathEffects.Normal()])
+			elif enrichment[i][y] == 0:
+				rects[i][y].set_linewidth(1)
+	
+	# Remove top/right/bottom axes
+	for ax in [ax1,ax2]:
+		ax.spines['top'].set_visible(False)
+		ax.spines['right'].set_visible(False)
+		ax.spines['bottom'].set_visible(False)
+		ax.tick_params(axis='x', which='both', bottom='on', top='off', labelbottom='on')
+		ax.tick_params(axis='y', which='both', left='on', right='off', labelleft='on')
+	
+	
+	
+	#ax1.legend(loc='center right', bbox_to_anchor=(1.2, 0),fancybox=True, shadow=True)
+	## Showing the plot
+	plt.tight_layout()
+	fig.subplots_adjust(wspace=0.2)
+	if pdf:
+		pdf.savefig()
+		plt.close()
+	elif svg:
+		if svg == True:
+			plt.savefig(counts_type+'.svg')
+		else :
+			if os.path.splitext(svg)[1] == '.svg':
+				plt.savefig('.'.join((os.path.splitext(svg)[0],counts_type,'svg')))
+			else:
+				plt.savefig('.'.join((svg,counts_type,'svg')))
+	elif png:
+		if png == True:
+			plt.savefig(counts_type+'.png')
+		else :
+			if os.path.splitext(png)[1] == '.png':
+				plt.savefig('.'.join((os.path.splitext(png)[0],counts_type,'png')))
+			else:
+				plt.savefig('.'.join((png,counts_type,'png')))
+	else:
+		plt.show()
+	## Save on disk it as a png image 
+	#fig.savefig('image_output.png', dpi=300, format='png', bbox_extra_artists=(lgd,), bbox_inches='tight')
+
+
+def filter_categs_on_biotype(selected_biotype,cpt) :
+	filtered_cpt = {}
+	for sample in cpt:
+		filtered_cpt[sample] = {}
+		for feature,count in cpt[sample].items():
+			if feature[1] == selected_biotype:
+				filtered_cpt[sample][feature[0]] = count
+	return filtered_cpt
+	
+
+##########################################################################
+#                           MAIN                                         #
+##########################################################################
+
+def usage_message(name=None):                                                            
+    return '''
+    Generate genome indexes:
+         python ALFA.py -a GTF_FILE  [-g GENOME_INDEX]
+                                         [--chr_len CHR_LENGTHS_FILE]
+    Process BAM file(s):
+         python ALFA.py -g GENOME_INDEX -i BAM1 LABEL1 [BAM2 LABEL2 ...]
+                                         [--bedgraph] [-s STRAND]
+                                         [-n] [--pdf output.pdf]
+                                         [-d {1,2,3,4}] [-t YMIN YMAX]
+    Index genome + process BAM:
+         python ALFA.py -a GTF_FILE [-g GENOME_INDEX]
+                            -i BAM1 LABEL1 [BAM2 LABEL2 ...]
+                            [--chr_len CHR_LENGTHS_FILE]
+                            [--bedgraph][-s STRAND]
+                            [-n] [--pdf output.pdf]
+                            [-d {1,2,3,4}] [-t YMIN YMAX]
+                            
+    Process previously created ALFA counts file(s):
+         python ALFA.py -c COUNTS1 [COUNTS2 ...]
+                            [-s STRAND]
+                            [-n] [--pdf output.pdf]
+                            [-d {1,2,3,4}] [-t YMIN YMAX]
+
+        '''
+
+
+#### Parse command line arguments and store them in 'options'
+parser = argparse.ArgumentParser(formatter_class=argparse.RawTextHelpFormatter, usage=usage_message())
+parser.add_argument('--version', action='version', version='version 1.0', help="show program's version number and exit\n\n-----------\n\n")
+# Options concernant l'index
+parser.add_argument('-g','--genome_index', help="Genome index files path and basename for existing index, or path and basename for new index creation\n\n")
+parser.add_argument('-a','--annotation', metavar = "GTF_FILE", help='Genomic annotations file (GTF format)\n\n')
+parser.add_argument('--chr_len', help='Tabulated file containing chromosome names and lengths\n\n-----------\n\n')
+
+# Options pour l'étape d'intersection
+parser.add_argument('-i','--input','--bam', dest='input', metavar=('BAM_FILE1 LABEL1',""), nargs='+', help='Input BAM file(s) and label(s). The BAM files must be sorted by position.\n\n')
+parser.add_argument('--bedgraph', action='store_const',default = False, const = True, help="Use this options if your input file(s) is(are) already in bedgraph format\n\n")
+parser.add_argument('-c','--counts',metavar=('COUNTS_FILE',""), nargs='+', help="Use this options instead of '-i/--input' to provide ALFA counts files as input \ninstead of bam/bedgraph files.\n\n")
+parser.add_argument('-s','--strandness', dest="strandness", nargs=1, action = 'store', default = ['unstranded'], choices = ['unstranded','forward','reverse','fr-firststrand','fr-secondstrand'], metavar="", help ="Library orientation. Choose within: 'unstranded', 'forward'/'fr-firststrand' \nor 'reverse'/'fr-secondstrand'. (Default: 'unstranded')\n\n-----------\n\n")
+
+# Options concernant le plot
+parser.add_argument('-biotype_filter',nargs=1,help=argparse.SUPPRESS)#"Make an extra plot of categories distribution using only counts of the specified biotype.")
+parser.add_argument('-d','--categories_depth', type=int, default='3', choices=range(1,5), help = "Use this option to set the hierarchical level that will be considered in the GTF file (default=3): \n(1) gene,intergenic; \n(2) intron,exon,intergenic; \n(3) 5'UTR,CDS,3'UTR,intron,intergenic; \n(4) start_codon,5'UTR,CDS,3'UTR,stop_codon,intron,intergenic. \n\n")
+parser.add_argument('--pdf', nargs='?', default=False, help="Save produced plots in PDF format at specified path ('categories_plots.pdf' if no argument provided)\n\n")
+parser.add_argument('--png', nargs='?', default=False, const=True, help="Save produced plots in PNG format with provided argument as basename \nor 'categories.png' and 'biotypes.png' if no argument provided\n\n")
+parser.add_argument('--svg', nargs='?', default=False, const=True, help="Save produced plots in SVG format with provided argument as basename \nor 'categories.svg' and 'biotypes.svg' if no argument provided\n\n")
+parser.add_argument('-n','--no_plot', dest='quiet', action='store_const', default=False, const=True, help="Do not show plots\n\n")
+parser.add_argument('-t','--threshold', dest='threshold', nargs = 2, metavar=("ymin","ymax"), type=float , help="Set axis limits for enrichment plots\n\n")
+
+if len(sys.argv)==1:
+    parser.print_usage()
+    sys.exit(1)
+    
+options = parser.parse_args()
+
+def required_arg(arg, aliases):
+	if not arg:
+		print >> sys.stderr, "\nError: %s argument is missing.\n" %aliases
+		parser.print_usage()
+		sys.exit()
+
+def check_files_enxtension(files):
+	return
+
+# Booleans for steps to be executed
+make_index = False
+intersect_reads = False
+process_counts = False
+
+#### Check arguments conformity and define which steps have to be perfomed
+if options.counts :
+	# Aucun autre argument requis
+	# Vérifier extension input
+	
+	# Action : Faire le plot uniquement
+	process_counts = True
+else:
+	if options.annotation :
+		# Vérifier si présence -gi
+		if options.genome_index :
+			genome_index_basename = options.genome_index
+		else:
+			genome_index_basename = options.annotation.split("/")[-1].split(".gtf")[0]
+		# Vérifier si un fichier existe déjà: 
+		if os.path.isfile(genome_index_basename+".stranded.index") :
+			if options.input:
+				print >> sys.stderr, "\nWarning: a index file named '%s' already exists and will be used. If you want to create a new index, please delete this file or specify an other path." %(genome_index_basename+".stranded.index")
+			else:
+				sys.exit("\nError: a index file named %s already exists. If you want to create a new index, please delete this file or specify an other path.\n" %(genome_index_basename+".stranded.index"))
+		# sinon -> action : index à faire
+		else :
+			make_index = True
+	# si l'index n'est pas  à faire :
+	if options.input:
+		#	Arguments requis: input, genome_index
+		if 'genome_index_basename' not in locals():
+			required_arg(options.genome_index, "-g/--genome_index")
+			genome_index_basename = options.genome_index
+		required_arg(options.input, "-i/--input/--bam")
+		for i in xrange(0, len(options.input), 2):
+			try :
+				extension = os.path.splitext(options.input[i+1])[1]
+				if extension == ".bam" or extension == '.bedgraph' or extension == '.bg':
+					sys.exit("Error: it seems input files and associated labels are not correctly provided.\n\
+					Make sure to follow the expected format : -i Input_file1 Label1 [Input_file2 Label2 ...].")
+			except:
+				sys.exit("Error: it seems input files and associated labels are not correctly provided.\nMake sure to follow the expected format : -i Input_file1 Label1 [Input_file2 Label2 ...].")
+
+		intersect_reads = True
+	# Vérifier input's extension
+	#TODO
+if not (options.counts or options.input or options.annotation):
+	sys.exit("\nError : some arguments are missing At least '-a', '-c' or '-i' is required. Please refer to help (-h/--help) and usage cases for more details.\n")
+if not options.counts:
+	# Declare genome_index variables
+	stranded_genome_index = genome_index_basename+".stranded.index"
+	unstranded_genome_index = genome_index_basename+".unstranded.index"
+	if options.strandness[0] == "unstranded":
+		genome_index = unstranded_genome_index
+	else:
+		genome_index = stranded_genome_index
+
+
+
+#### Initialization of some variables
+
+# Initializing the category priority order, coding biotypes and the final list
+prios = {'start_codon': 7, 'stop_codon': 7, 'five_prime_utr': 6, 'three_prime_utr': 6, 'UTR': 6, 'CDS': 5, 'exon': 4, 'transcript': 3, 'gene': 2, 'antisense': 1,'intergenic': 0}
+
+biotype_prios = None
+#biotype_prios = {"protein_coding":1, "miRNA":2}
+
+# Possibles groups of categories to plot
+categs_group1={'start': ['start_codon'], '5UTR': ['five_prime_utr','UTR'], 'CDS': ['CDS', 'exon'], '3UTR': ['three_prime_utr'], 'stop': ['stop_codon'], 'introns': ['transcript', 'gene'], 'intergenic': ['intergenic'], 'antisense': ['antisense']}
+categs_group2={'5UTR': ['five_prime_utr', 'UTR'], 'CDS': ['CDS', 'exon','start_codon','stop_codon'], '3UTR': ['three_prime_utr'], 'introns': ['transcript', 'gene'], 'intergenic': ['intergenic'], 'antisense': ['antisense']}
+categs_group3={'exons': ['five_prime_utr', 'three_prime_utr', 'UTR', 'CDS', 'exon','start_codon','stop_codon'], 'introns': ['transcript', 'gene'], 'intergenic': ['intergenic'], 'antisense': ['antisense']}
+categs_group4={'gene': ['five_prime_utr', 'three_prime_utr', 'UTR', 'CDS', 'exon','start_codon','stop_codon', 'transcript', 'gene'], 'intergenic': ['intergenic'], 'antisense': ['antisense']}
+categs_groups = [categs_group4,categs_group3,categs_group2,categs_group1] # Order and merging for the final plot
+cat_list = ['5UTR', 'start', 'CDS', 'stop', '3UTR', 'exons', 'introns', 'gene', 'intergenic', 'antisense']
+
+
+# biotypes list
+biotypes ={'protein_coding','polymorphic_pseudogene','TR_C_gene','TR_D_gene','TR_J_gene','TR_V_gene','IG_C_gene','IG_D_gene','IG_J_gene','IG_V_gene',"3prime_overlapping_ncrna","lincRNA","macro_lncRNA","miRNA","misc_RNA","Mt_rRNA","Mt_tRNA","processed_transcript","ribozyme","rRNA","scaRNA","sense_intronic","sense_overlapping","snoRNA","snRNA","sRNA","TEC","vaultRNA","antisense","transcribed_processed_pseudogene","transcribed_unitary_pseudogene","transcribed_unprocessed_pseudogene","translated_unprocessed_pseudogene","TR_J_pseudogene","TR_V_pseudogene","unitary_pseudogene","unprocessed_pseudogene","processed_pseudogene","IG_C_pseudogene","IG_J_pseudogene","IG_V_pseudogene","pseudogene","ncRNA","tRNA"} # Type: set (to access quickly)
+
+# Grouping of biotypes:
+biotypes_group1={'protein_coding':['protein_coding'],'pseudogenes':['polymorphic_pseudogene',"transcribed_processed_pseudogene","transcribed_unitary_pseudogene","transcribed_unprocessed_pseudogene","translated_unprocessed_pseudogene","TR_J_pseudogene","TR_V_pseudogene","unitary_pseudogene","unprocessed_pseudogene","processed_pseudogene","IG_C_pseudogene","IG_J_pseudogene","IG_V_pseudogene","pseudogene"],'TR':['TR_C_gene','TR_D_gene','TR_J_gene','TR_V_gene'],'IG':['IG_C_gene','IG_D_gene','IG_J_gene','IG_V_gene'],\
+	'MT_RNA':["Mt_rRNA","Mt_tRNA"],\
+	'ncRNA':["lincRNA","macro_lncRNA","3prime_overlapping_ncrna","ncRNA"],\
+	"others":["misc_RNA","processed_transcript","ribozyme","scaRNA","sense_intronic","sense_overlapping","TEC","vaultRNA"],
+	"antisense":["antisense"]}
+for biot in ["miRNA","snoRNA","snRNA","rRNA","sRNA","tRNA"]:
+	biotypes_group1[biot]=[biot]
+
+# # Initializing the unkown features lits
+unknown_feature = []
+
+# Initializing the genome category counter dict
+cpt_genome = {}
+
+
+if process_counts :
+	#### If input files are the categories counts, just load them and continue to recategorization step
+	cpt,cpt_genome,samples_names = read_counts_files(options.counts)
+else:
+	#### Create genome index if needed and get the sizes of categories
+	if make_index :
+		#### Get the chromosome lengths
+		lengths = get_chromosome_lengths(options)
+		# Generating the genome index files if the user didn't provide them
+		create_genome_index(options.annotation, unstranded_genome_index, stranded_genome_index, cpt_genome, prios, biotypes, lengths)
+
+
+	#print '\nChr lengths:', lengths
+
+if intersect_reads:
+	# If the indexes already exist, read them to compute the sizes of the categories in the genome and retrieve the chromosome lengths
+	if not make_index :
+		print "\n### Reading genome indexes\n...\r",
+		sys.stdout.flush()
+	lengths={}
+	with open(genome_index, 'r') as genome_index_file:
+		for line in genome_index_file:
+			if line[0] == "#":
+				lengths[line.split('\t')[0][1:]] = int(line.split('\t')[1])
+			else :
+				add_info(cpt_genome, line.rstrip().split('\t')[4:], line.split('\t')[1], line.split('\t')[2], biotype_prios = None, categ_prios = prios)
+						
+	#### Computing the genome intergenic count: sum of the chr lengths minus sum of the genome annotated intervals
+	cpt_genome[('intergenic','intergenic')] = sum(lengths.itervalues()) - sum([v for x,v in cpt_genome.iteritems() if x != ('antisense','antisense')])
+	if not make_index :
+		print "Done!"
+	#print '\nGenome category counts:'
+	#for key,val in cpt_genome.iteritems():
+		#print key,"\t",val
+
+
+	#### Create the Bedgraph files if needed and get the files list
+
+	if not options.bedgraph:
+		# Generating the BEDGRAPH files is the user provided BAM file(s) and get the samples labels (this names will be used in the plot legend)
+		samples_files, samples_names = create_bedgraph_files(options.input,options.strandness[0])
+	else:
+		# Just initialize the files list with the bedgraph paths
+		samples_files = [options.input[i] for i in range(0,len(options.input),2)]
+		# and get the labels
+		samples_names = [options.input[i] for i in range(1,len(options.input),2)]
+	#### Retrieving chromosome names saved in index
+	chrom_list = get_chromosome_names_in_index(genome_index)
+	#### Processing the BEDGRAPH files: intersecting the bedgraph with the genome index and count the number of aligned positions in each category
+	cpt = intersect_bedgraphs_and_index_to_counts_categories(samples_files,samples_names,prios,genome_index, options.strandness[0], biotype_prios = None)
+
+	#### Write the counts on disk
+	write_counts_in_files(cpt,cpt_genome)
+
+if not (intersect_reads or process_counts) or (options.quiet and options.pdf == False):
+	quit("\n### End of program")
+print "\n### Generating plots"
+# Updating the biotypes lists (biotypes and 'biotype_group1'): adding the 'unknow biotypes' found in gtf/index
+if unknown_feature == []: # 'unknown_feature' is define only during the index generation
+	# Browse the feature to determine whether some biotypes are 'unknown'
+	for sample,counts in cpt.items():
+		for (cat,biot) in counts:
+			if biot not in biotypes and cat not in unknown_feature:
+				unknown_feature.append(biot)
+for new_biot in unknown_feature:
+	biotypes.add(new_biot)
+	biotypes_group1["others"].append(new_biot)
+biotypes = sorted(biotypes)
+# move antisense categ to the end of the list
+biotypes.remove('antisense')
+biotypes.append('antisense')
+biotypes_group1 = sorted(biotypes_group1)
+
+
+#print '\nCounts for every category/biotype pair: ',cpt
+
+# Generating plots
+if options.pdf != False:
+	if options.pdf == None:
+		options.pdf = "categories_plots.pdf"
+	pdf = PdfPages(options.pdf)
+else:
+	pdf = False
+
+selected_biotype = None
+if options.biotype_filter:
+	options.biotype_filter = options.biotype_filter[0]
+	for sample in cpt:
+		for feature in cpt[sample]:
+			biotype = feature[1]
+			if options.biotype_filter.lower() == biotype.lower():
+				selected_biotype=biotype
+				break
+	if selected_biotype == None :
+		print "\nError: biotype '"+options.biotype_filter+"' not found. Please check the biotype name and that this biotype exists in your sample(s)."
+		sys.exit()
+
+#Print a warning message if the UTRs are not specified as 5' or 3' (they will be ploted as 5'UTR)
+if 'UTR'  in [categ[0] for counts in cpt.values() for categ in counts.keys()]:
+	print '''\nWARNING: (some) 5'UTR/3'UTR are not precisely defined. Consequently, positions annotated as "UTR" will be counted as "5'UTR"\n'''
+
+#### Make the plot by categories
+	#### Recategorizing with the final categories
+final_cats=categs_groups[options.categories_depth-1]
+final_cat_cpt,final_genome_cpt, filtered_cat_cpt = group_counts_by_categ(cpt,cpt_genome,final_cats,selected_biotype)
+	#### Display the distribution of specified categories (or biotypes) in samples on a barplot
+# Remove the 'antisense' category if the library type is 'unstranded'
+for dic in cpt.values():
+	if ('antisense','antisense') in dic.keys(): break
+else:
+	cat_list.remove('antisense')
+make_plot(cat_list,samples_names,final_cat_cpt,final_genome_cpt,pdf, "categories",options.threshold, svg = options.svg, png = options.png)
+if selected_biotype :
+	make_plot(cat_list,samples_names,filtered_cat_cpt,final_genome_cpt,pdf, "categories",options.threshold,title="Categories distribution for '"+selected_biotype+"' biotype", svg = options.svg, png = options.png)
+
+#### Make the plot by biotypes
+	#### Recategorizing with the final categories
+final_cat_cpt,final_genome_cpt = group_counts_by_biotype(cpt,cpt_genome,biotypes)
+	#### Display the distribution of specified categories (or biotypes) in samples on a barplot
+make_plot(biotypes,samples_names,final_cat_cpt,final_genome_cpt,pdf, "biotypes",options.threshold, svg = options.svg, png = options.png)
+
+
+
+	##### Recategorizing with the final categories
+#final_cat_cpt,final_genome_cpt = group_counts_by_biotype(cpt,cpt_genome,biotypes_group1)
+	##### Display the distribution of specified categories (or biotypes) in samples on a barplot
+#make_plot(biotypes_group1,samples_names,final_cat_cpt,final_genome_cpt,pdf,"Biotype groups", options.threshold, title="Biotypes distribution in mapped reads \n(biotypes are grouped by 'family')", svg = options.svg, png = options.png)
+
+
+if options.pdf:
+	pdf.close()
+	print "\n### Plots saved in pdf file: %s" %options.pdf
+	
+print "\n### End of program"
\ No newline at end of file
--- a/data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.py	Thu Oct 27 05:45:31 2016 -0400
+++ b/data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.py	Thu Oct 27 06:49:58 2016 -0400
@@ -156,7 +156,7 @@
 def generate_alfa_indexes(path_to_alfa, gtf_file_name):
     logging.info("____________________________________________________________")
     logging.info("*** Generating alfa indexes from %s" % gtf_file_name)
-    alfa_result = subprocess.Popen(['python', path_to_alfa + 'ALFA.py', '-a',  gtf_file_name], stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+    alfa_result = subprocess.Popen(['python', path_to_alfa, '-a',  gtf_file_name], stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
     alfa_out, alfa_err = alfa_result.communicate()
     if alfa_err != None and not re.search('### End of program', alfa_err):
         msg = 'Generation Failed due an alfa error: %s' % (alfa_err)
@@ -181,8 +181,9 @@
 def main():
     options, args = get_arg()
     galaxy_root_dir = args[0]
+    tool_dir = args[1]
 
-    path_to_alfa = os.path.join(galaxy_root_dir, 'tools/alfa/')
+    path_to_alfa = os.path.join(tool_dir, 'ALFA.py')
     path_to_tmp_dir = os.path.join(galaxy_root_dir, 'database/tmp/')
 
     if options.output_filename == None:
--- a/data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.xml	Thu Oct 27 05:45:31 2016 -0400
+++ b/data_manager_build_alfa_indexes/data_manager/data_manager_build_alfa_indexes.xml	Thu Oct 27 06:49:58 2016 -0400
@@ -5,7 +5,7 @@
   	 <requirement type="package" version="0.1.0">alfa</requirement>
   </requirements>
 
-  <command interpreter="python">data_manager_build_alfa_indexes.py -e "${reference_source['kingdom']}" "${reference_source['species_name']}" -o "${out_file}" $__root_dir__ </command>
+  <command interpreter="python">data_manager_build_alfa_indexes.py -e "${reference_source['kingdom']}" "${reference_source['species_name']}" -o "${out_file}" $__root_dir__ $__tool_directory__</command>
 
   <inputs>
     <conditional name="reference_source">
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/data_manager_build_alfa_indexes/tool-data/alfa_indexes.loc.sample~	Thu Oct 27 06:49:58 2016 -0400
@@ -0,0 +1,5 @@
+#<species>	<version>	<release>	<value>	<dbkey>	<name>	<prefix>
+#Arabidopsis_thaliana	TAIR10	30	Arabidopsis_t_TAIR10_30	Arabidopsis_t_TAIR10_30	Arabidopsis thaliana: TAIR10	<path_to_alfa_indexes>/Arabidopsis_thaliana.TAIR10.30
+#Drosophila_melanogaster	dm6	30	Drosophila_m_dm6_30	Drosophila_m_dm6_30	Drosophila melanogaster: dm6	<path_to_alfa_indexes>/Drosophila_melanogaster.BDGP6.30
+#Homo_sapiens	v38	82	Homo_s_v38_82	Homo_s_v38_82	Homo sapiens: v38	<path_to_alfa_indexes>/Homo_sapiens.GRCh38.82
+#Mus_musculus	v38	83	Mus_m_v38_83	Mus_m_v38_83	Mus musculus: v38 	<path_to_alfa_indexes>/Mus_musculus.GRCm38.83.chr
--- a/data_manager_build_alfa_indexes/tool_dependencies.xml	Thu Oct 27 05:45:31 2016 -0400
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,5 +0,0 @@
-<tool_dependency>
-    <package name="alfa" version="0.1.0">
-        <repository toolshed="https://toolshed.g2.bx.psu.edu" name="alfa" owner="charles-bernard" changeset_revision="c7279df3f7ba" />
-    </package>
-</tool_dependency>
\ No newline at end of file