import sys, collections, itertools, os.path

if len( sys.argv ) != 3:
   sys.stderr.write( "Script to prepare annotation for DEXSeq.\n\n" )
   sys.stderr.write( "Usage: python %s <in.gtf> <out.gff>\n\n" % os.path.basename(sys.argv[0]) )
   sys.stderr.write( "This script takes an annotation file in Ensembl GTF format\n" )
   sys.stderr.write( "and outputs a 'flattened' annotation file suitable for use\n" )
   sys.stderr.write( "with the count_in_exons.py script.\n" )
   sys.exit(1)

try:
   import HTSeq
except ImportError:
   sys.stderr.write( "Could not import HTSeq. Please install the HTSeq Python framework\n" )   
   sys.stderr.write( "available from http://www-huber.embl.de/users/anders/HTSeq\n" )   
   sys.exit(1)

gtf_file = sys.argv[1]
out_file = sys.argv[2]

## make sure that it can handle GFF files.
parent_child_map = dict()
for feature in HTSeq.GFF_Reader( gtf_file ):
   if feature.type in ['mRNA', 
      'transcript', 
      'ncRNA', 
      'miRNA', 
      'pseudogenic_transcript', 
      'rRNA', 
      'snoRNA', 
      'snRNA', 
      'tRNA', 
      'scRNA']:
      parent_child_map[feature.attr['ID']] = feature.attr['Parent']

# Step 1: Store all exons with their gene and transcript ID 
# in a GenomicArrayOfSets

exons = HTSeq.GenomicArrayOfSets( "auto", stranded=True )
for f in HTSeq.GFF_Reader( gtf_file ):
   if not f.type in ["exon", "pseudogenic_exon"]:
      continue
   if not f.attr.get('gene_id'):
      f.attr['gene_id'] = parent_child_map[f.attr['Parent']]
      f.attr['transcript_id'] = f.attr['Parent']
   f.attr['gene_id'] = f.attr['gene_id'].replace( ":", "_" )
   exons[f.iv] += ( f.attr['gene_id'], f.attr['transcript_id'] )

# Step 2: Form sets of overlapping genes

# We produce the dict 'gene_sets', whose values are sets of gene IDs. Each set
# contains IDs of genes that overlap, i.e., share bases (on the same strand).
# The keys of 'gene_sets' are the IDs of all genes, and each key refers to
# the set that contains the gene.
# Each gene set forms an 'aggregate gene'.

gene_sets = collections.defaultdict( lambda: set() )
for iv, s in exons.steps():
   # For each step, make a set, 'full_set' of all the gene IDs occuring
   # in the present step, and also add all those gene IDs, whch have been
   # seen earlier to co-occur with each of the currently present gene IDs.
   full_set = set()
   for gene_id, transcript_id in s:
      full_set.add( gene_id )
      full_set |= gene_sets[ gene_id ]
   # Make sure that all genes that are now in full_set get associated
   # with full_set, i.e., get to know about their new partners
   for gene_id in full_set:
      assert gene_sets[ gene_id ] <= full_set
      gene_sets[ gene_id ] = full_set


# Step 3: Go through the steps again to get the exonic sections. Each step
# becomes an 'exonic part'. The exonic part is associated with an
# aggregate gene, i.e., a gene set as determined in the previous step, 
# and a transcript set, containing all transcripts that occur in the step.
# The results are stored in the dict 'aggregates', which contains, for each
# aggregate ID, a list of all its exonic_part features.

aggregates = collections.defaultdict( lambda: list() )
for iv, s in exons.steps( ):
   # Skip empty steps
   if len(s) == 0:
      continue
   # Take one of the gene IDs, find the others via gene sets, and
   # form the aggregate ID from all of them   
   gene_id = list(s)[0][0]
   assert set( gene_id for gene_id, transcript_id in s ) <= gene_sets[ gene_id ] 
   aggregate_id = '+'.join( gene_sets[ gene_id ] )
   # Make the feature and store it in 'aggregates'
   f = HTSeq.GenomicFeature( aggregate_id, "exonic_part", iv )   
   f.source = os.path.basename( sys.argv[1] )
   f.attr = {}
   f.attr[ 'gene_id' ] = aggregate_id
   transcript_set = set( ( transcript_id for gene_id, transcript_id in s ) )
   f.attr[ 'transcripts' ] = '+'.join( transcript_set )
   aggregates[ aggregate_id ].append( f )


# Step 4: For each aggregate, number the exonic parts

aggregate_features = []
for l in aggregates.values():
   for i in xrange( len(l)-1 ):
      assert l[i].name == l[i+1].name, str(l[i+1]) + " has wrong name"
      assert l[i].iv.end <= l[i+1].iv.start, str(l[i+1]) + " starts too early"
      if l[i].iv.chrom != l[i+1].iv.chrom:
         raise ValueError, "Same name found on two chromosomes: %s, %s" % ( str(l[i]), str(l[i+1]) )
      if l[i].iv.strand != l[i+1].iv.strand:
         raise ValueError, "Same name found on two strands: %s, %s" % ( str(l[i]), str(l[i+1]) )
   aggr_feat = HTSeq.GenomicFeature( l[0].name, "aggregate_gene", 
      HTSeq.GenomicInterval( l[0].iv.chrom, l[0].iv.start, 
         l[-1].iv.end, l[0].iv.strand ) )
   aggr_feat.source = os.path.basename( sys.argv[1] )
   aggr_feat.attr = { 'gene_id': aggr_feat.name }
   for i in xrange( len(l) ):
      l[i].attr['exonic_part_number'] = "%03d" % ( i+1 )
   aggregate_features.append( aggr_feat )
      
      
# Step 5: Sort the aggregates, then write everything out

aggregate_features.sort( key = lambda f: ( f.iv.chrom, f.iv.start ) )

fout = open( out_file, "w" ) 
for aggr_feat in aggregate_features:
   fout.write( aggr_feat.get_gff_line() )
   for f in aggregates[ aggr_feat.name ]:
      fout.write( f.get_gff_line() )

fout.close()