GTF2psix.py

import numpy as np
import pandas as pd
import argparse
from tqdm import tqdm
import gzip
from gtfparse import read_gtf


'''
Process GTF annotation to create an index of cassette exons and constitutive introns compatible with Psix.
Author: Carlos F Buen Abad Najar
cfbuenabadn [at] berkeley [dot] edu
'''

description = 'Create cassette exon annotation compatible with Psix, from a GTF file.'
description += ' Usage: python GTF2psix.py --gtf <path/to/gtf> [optional arguments]'

parser = argparse.ArgumentParser(description=description)

parser.add_argument('--gtf', type=str, required=True, help='GTF file to convert.')

parser.add_argument('-o', '--output_name', type=str, required=False, default='psix_annotation', help='Name for output file. Psix will add a .tab.gz extension')

parser.add_argument('-g', '--gene_type', type=str, required=False, default='all', help='Type of genes to include in annotation. E.g.: protein_coding.')

parser.add_argument('-n', '--gene_name', type=str, required=False, default='gene_id', help='Which gene identifier should be used to create the annotation. Use "gene_id" (if available in your gtf file) to create annotation using ensembl IDs. Use "gene_name" to create annotation with gene symbols. If none of these tags are present in your GTF file, you should indicate the tag in the GTF that you want to use.')

parser.add_argument('-trim', '--no_trim_id', action='store_true', required=False, help='Indicate if the gene ID should be trimmed. Some annotations (such as Gencode) have the following format for their ensembl id: "ENS0000001234.1". Passing this argument will trim the id in the annotation to "ENS0000001234". The default is to trim the annotation. If the annotation does not need trimming, it will not have an effect.')

parser.add_argument('-gt', '--gene_type_tag', type=str, required=False, default='gene_type', help='Indicate the tag to be used for gene type. By default we use "gene_type". Some annotations use "gene_biotype".')

parser.add_argument('-tt', '--transcript_type_tag', type=str, required=False, default='transcript_type', help='Indicate the tag to be used for transcript type. By default we use "transcript_type". Some annotations use "transcript_biotype".')

parser.add_argument('-e', '--exclude_chromosome', type=str, required=False, default='', help='Chromosomes to exclude, separated by comma. E.g.: chrX,chrY,chrM')

def process_gtf(gtf_file, exclude, gene_name, no_trim_id, gene_type_tag, transcript_type_tag):

    print('Processing GTF file...')
    
    exclude_chromosomes = exclude.split(',')
    
    gtf = read_gtf(gtf_file)
    
    try:
    
        try:
            cols = ['seqname', 'start', 'end', 'strand', 'transcript_id', gene_type_tag, gene_name, transcript_type_tag]
            gtf = gtf.loc[pd.Series([x not in exclude_chromosomes for x in gtf.seqname]) & (gtf.feature == 'exon'), cols]
            gtf.columns = ['chrom', 'start', 'end', 'strand', 'transcript', 'gene_type', 'gene', 'transcript_type']
        except:
            cols = ['seqname', 'start', 'end', 'strand', 'transcript_id', gene_name]
            gtf = gtf.loc[pd.Series([x not in exclude_chromosomes for x in gtf.seqname]) & (gtf.feature == 'exon'), cols]
            gtf.columns = ['chrom', 'start', 'end', 'strand', 'transcript', 'gene']
            
    except:
        raise Exception('Isufficient information to create annotation. transcript_id is needed to find cassette exons.')
    
    if not no_trim_id:
        gtf['gene'] = gtf.gene.str.replace(r'\.\d+$', '', regex=True)
    
    gtf.transcript = [x.split('.')[0] for x in gtf.transcript]
    
    
    print('Finished processing GTF file')

    gtf = gtf.loc[gtf.gene != '']
    
    return gtf


def get_dirs(gtf_gene, gene):
    donor_dir = {}
    acceptor_dir = {}

    old_transcript = ''
    for idx, row in gtf_gene.sort_values(['transcript', 'start', 'end']).iterrows():
        transcript = row.transcript

        if old_transcript != transcript:

            counter = 1
            old_transcript = transcript

        else:

            transcript_e = transcript + '.' + str(counter)
            acceptor = str(int(row.start)-1)

            if not acceptor in acceptor_dir.keys():
                acceptor_dir.update({acceptor:{donor:[transcript_e]}})

            elif not donor in acceptor_dir[acceptor].keys():
                acceptor_dir[acceptor].update({donor:[transcript_e]})

            else:
                acceptor_dir[acceptor][donor].append(transcript_e)


            if acceptor in donor_dir[donor].keys():
                donor_dir[donor][acceptor].append(transcript_e)

            else:
                donor_dir[donor].update({acceptor:[transcript_e]})

            counter += 1

        donor = str(int(row.end)+1)

        if not donor in donor_dir.keys():
            #print('repetido')
            donor_dir.update({donor:{}})

    return donor_dir, acceptor_dir

    
def get_const_introns(fh, gtf):
    
    try:
        pc_gtf = gtf.loc[(gtf.transcript_type == 'protein_coding') & (gtf.gene_type == 'protein_coding')]
    except:
        pc_gtf = gtf
    
    for i in tqdm(pc_gtf.groupby('gene'), leave=True, position=0):
        t_gtf = i[1]
        transcripts = t_gtf.transcript.unique()

        donor_dir, acceptor_dir = get_dirs(t_gtf, transcripts[0])

        introns_list = []
        for donor in donor_dir:

            if len(donor_dir[donor].keys()) > 0:

                acceptor = list(donor_dir[donor].keys())[0]

                intron = str(t_gtf.chrom.unique()[0])
                intron += ':' + str(donor)
                intron += ':' + str(acceptor)
                intron += ':' + str(t_gtf.strand.unique()[0])

                introns_list.append(intron)

        for t in transcripts:

            t_gtf = i[1].loc[i[1].transcript == t]

            introns = []

            donor_dir, acceptor_dir = get_dirs(t_gtf, transcripts[0])

            for donor in donor_dir:

                if len(donor_dir[donor].keys()) > 0:

                    acceptor = list(donor_dir[donor].keys())[0]

                    intron = str(t_gtf.chrom.unique()[0])
                    intron += ':' + str(donor)
                    intron += ':' + str(acceptor)
                    intron += ':' + str(t_gtf.strand.unique()[0])

                    introns.append(intron)

            introns_list = [x for x in introns_list if x in introns]

        intron_counts = 1
        if len(introns_list) > 0:
            for intron in introns_list:
                chrom, start, end, strand = intron.split(':')

                intron_number = i[0] + '_' + str(intron_counts)
                intron_name = intron_number + '_CI'
                intron_loc = chrom + ':' + start + '-' + end + ':' + strand
                
                linea = '\t'.join([intron_name, intron_loc, intron_number, i[0]]) + '\n'
                
                fh.write(linea)

                intron_counts += 1

    fh.close()
    
def get_cassette_exons(gtf_gene, donor_dir, acceptor_dir):

    ase_dir = {}

    for donor in donor_dir.keys():                              # Parse all donors
        if len(donor_dir[donor].keys()) > 1:                    # Check for donors with more than one acceptor
            acceptors = sorted(donor_dir[donor].keys())         # Sort them, so that we go in order for acceptors

            for n in range(1, len(acceptors)):                     # start with the 2nd acceptor; we assume no exon inside 1st intron

                i = acceptors[n]                                   # parse through acceptors
                intron_list = donor_dir[donor][i]                  # introns that end in the nth acceptor
                for intron in intron_list:                      # parsing through all the introns

                    for j in acceptor_dir[i].keys():            # for each donor of the nth acceptor

                        if j == donor:
                            skipped_isoforms = [x for x in donor_dir[j][i] if x in acceptor_dir[i][j]] 

                        if j != donor:                          # check one that is different

                            cassette_isoforms = []
                            for acceptor_isoform in acceptor_dir[i][j]:
                                split_intron = acceptor_isoform.split('.')



                                upstream_intron = split_intron[0] + '.' + str(int(split_intron[1]) - 1)

                                for acceptor_final in donor_dir[donor].keys():

                                    if upstream_intron in donor_dir[donor][acceptor_final]:

                                        I1 = donor + ':' + acceptor_final
                                        I2 = j + ':' + i
                                        SE = donor + ':' + i

                                        evento = (I1, I2, SE)

                                        cassette_isoforms.append(upstream_intron + ':' + acceptor_isoform)

                                        nmd = upstream_intron.split('.')[0]
                                        
                                        try:

                                            nmd_type = gtf_gene.loc[gtf_gene.transcript == nmd].transcript_type.unique()[0]
                                            
                                        except:
                                            nmd_type = 'notype'

                                        event_key = '|'.join(evento)

                                        if event_key not in ase_dir.keys():

                                            ase_dir.update({event_key:[nmd_type]})

                                        else:
                                            ase_dir[event_key].append(nmd_type)

    return ase_dir


def process_gene_annotation(gtf, gene):
    
    gtf_gene = gtf.loc[gtf.gene == gene]
    
    donor_dir, acceptor_dir = get_dirs(gtf_gene, gene)
    
    ase_dir = get_cassette_exons(gtf_gene, donor_dir, acceptor_dir)
    
    chrom = list(gtf_gene.chrom)[0]
    strand = list(gtf_gene.strand)[0]
    
    event_sorted = sorted(ase_dir.keys())
    
    gene_events = ''

    notype_counts = 1
    pc_counts = 1
    nmd_counts = 1
    other_counts = 1
    
    for event in event_sorted:
        if 'protein_coding' in ase_dir[event]:
            event_name = gene + '_ProteinCoding_' + str(pc_counts)
            pc_counts += 1
        elif 'nonsense_mediated_decay' in ase_dir[event]:
            event_name = gene + '_NMD_' + str(nmd_counts)
            nmd_counts += 1
        ### modify the script here if you're interested in other tags
        ### elif 'your_tag' in ase_dir[event]:
        ###    event_name = gene + '_YOURTAG_' + str(yourtag_counts)
        elif 'notype' in ase_dir[event]:
            event_name = gene + '_UNKNOWNTYPE_' + str(notype_counts)
            notype_counts += 1
        else:
            event_name = gene + '_other_' + str(other_counts)
            other_counts += 1
            
        i1_start, i1_end = event.split('|')[0].split(':')
        intron_loc = chrom + ':' + i1_start + '-' + i1_end + ':' + strand
        i1_line = '\t'.join([event_name + '_I1', intron_loc, event_name, gene]) + '\n'
        gene_events += i1_line
        
        
        i2_start, i2_end = event.split('|')[1].split(':')
        intron_loc = chrom + ':' + i2_start + '-' + i2_end + ':' + strand
        i2_line = '\t'.join([event_name + '_I2', intron_loc, event_name, gene]) + '\n'
        gene_events += i2_line
        
        se_start, se_end = event.split('|')[2].split(':')
        intron_loc = chrom + ':' + se_start + '-' + se_end + ':' + strand
        se_line = '\t'.join([event_name + '_SE', intron_loc, event_name, gene]) + '\n'
        gene_events += se_line
        

    return gene_events


def write_annotation(gtf, out_file, gene_type = 'protein_coding'):
    
    if gene_type != 'all':
        print('Working on ' + gene_type + ' genes')
        
        try:
            gene_list = gtf.loc[gtf.gene_type == gene_type].gene.unique()
        except:
            raise Exception(
                'Gene type not found in GTF file. Please make sure that gene_type annotation exists, or use "--gene_type all" instead.'
            )
        
    else:
        print('Working on all genes')
        gene_list = gtf.gene.unique()
        
    fh = gzip.open(out_file + '.tab.gz', 'wt')
    
    fh.write('\t'.join(['', 'intron', 'event', 'gene']) + '\n')
    
    gene_counts = 1
    total_genes = len(gene_list)
    
    print('Working in a total of ' + str(total_genes))
    
    for gene in tqdm(gene_list, leave=True, position=0):
        rows = process_gene_annotation(gtf, gene)
        
        fh.write(rows)
            
        gene_counts += 1
        
    print('Processed ' + str(gene_counts-1)+'/'+str(total_genes))
    
    print('Getting constitutive introns')
    get_const_introns(fh, gtf)
    
    fh.close()
    print('Finished writing annotation')

if __name__ == '__main__':
    args = parser.parse_args()
    gtf_file = args.gtf
    output_name = args.output_name
    gene_type = args.gene_type
    gene_name = args.gene_name
    no_trim_id = args.no_trim_id
    gene_type_tag = args.gene_type_tag
    transcript_type_tag = args.transcript_type_tag
    exclude = args.exclude_chromosome
        
    gtf = process_gtf(gtf_file, exclude, gene_name, no_trim_id, gene_type_tag, transcript_type_tag)
    
    write_annotation(gtf, output_name, gene_type)