sashimi-plot.py

#!/usr/bin/env python

# Import modules
from argparse import ArgumentParser
import subprocess as sp
import sys, re, copy, os
from collections import OrderedDict

def define_options():
        # Argument parsing
        parser = ArgumentParser(description='Create sashimi plot for a given genomic region')
        parser.add_argument("-b", "--bam", type=str,
                help="""
                Individual bam file or file with a list of bam files.
                In the case of a list of files the format is tsv:
                1col: id for bam file,
                2col: path of bam file,
                3+col: additional columns
                """)
        parser.add_argument("-c", "--coordinates", type=str,
                help="Genomic region. Format: chr:start-end (1-based)")
        parser.add_argument("-o", "--out-prefix", type=str, dest="out_prefix", default="sashimi",
                help="Prefix for plot file name [default=%(default)s]")
        parser.add_argument("-S", "--out-strand", type=str, dest="out_strand", default="both",
                help="Only for --strand other than 'NONE'. Choose which signal strand to plot: <both> <plus> <minus> [default=%(default)s]")
        parser.add_argument("-M", "--min-coverage", type=int, default=1, dest="min_coverage",
                help="Minimum number of reads supporting a junction to be drawn [default=1]")
        parser.add_argument("-j", "--junctions-bed", type=str, dest = "junctions_bed", default="",
                help="Junction BED file name [default=no junction file]")
        parser.add_argument("-g", "--gtf",
                help="Gtf file with annotation (only exons is enough)")
        parser.add_argument("-s", "--strand", default="NONE", type=str,
                help="Strand specificity: <NONE> <SENSE> <ANTISENSE> <MATE1_SENSE> <MATE2_SENSE> [default=%(default)s]")
        parser.add_argument("--shrink", action="store_true",
                help="Shrink the junctions by a factor for nicer display [default=%(default)s]")
        parser.add_argument("-O", "--overlay", type=int,
                help="Index of column with overlay levels (1-based)")
        parser.add_argument("-A", "--aggr", type=str, default="",
                help="""Aggregate function for overlay: <mean> <median> <mean_j> <median_j>.
                        Use mean_j | median_j to keep density overlay but aggregate junction counts [default=no aggregation]""")
        parser.add_argument("-C", "--color-factor", type=int, dest="color_factor",
                help="Index of column with color levels (1-based)")
        parser.add_argument("--alpha", type=float, default=0.5,
                help="Transparency level for density histogram [default=%(default)s]")
        parser.add_argument("-P", "--palette", type=str,
                help="Color palette file. tsv file with >=1 columns, where the color is the first column")
        parser.add_argument("-L", "--labels", type=int, dest="labels", default=1,
                help="Index of column with labels (1-based) [default=%(default)s]")
        parser.add_argument("--height", type=float, default=2,
                help="Height of the individual signal plot in inches [default=%(default)s]")
        parser.add_argument("--ann-height", type=float, default=1.5, dest="ann_height",
                help="Height of annotation plot in inches [default=%(default)s]")
        parser.add_argument("--width", type=float, default=10,
                help="Width of the plot in inches [default=%(default)s]")
        parser.add_argument("--base-size", type=float, default=14, dest="base_size",
                help="Base font size of the plot in pch [default=%(default)s]")
        parser.add_argument("-F", "--out-format", type=str, default="pdf", dest="out_format",
                help="Output file format: <pdf> <svg> <png> <jpeg> <tiff> [default=%(default)s]")
        parser.add_argument("-R", "--out-resolution", type=int, default=300, dest="out_resolution",
                help="Output file resolution in PPI (pixels per inch). Applies only to raster output formats [default=%(default)s]")
#       parser.add_argument("-s", "--smooth", action="store_true", default=False, help="Smooth the signal histogram")
        return parser



def parse_coordinates(c):
        c = c.replace(",", "")
        chr = c.split(":")[0]
        start, end = c.split(":")[1].split("-")
        # Convert to 0-based
        start, end = int(start) - 1, int(end)
        return chr, start, end



def count_operator(CIGAR_op, CIGAR_len, pos, start, end, a, junctions):

        # Match
        if CIGAR_op == "M":
                for i in range(pos, pos + CIGAR_len):
                        if i < start or i >= end:
                                continue
                        ind = i - start
                        a[ind] += 1

        # Insertion or Soft-clip
        if CIGAR_op == "I" or CIGAR_op == "S":
                return pos

        # Deletion
        if CIGAR_op == "D":
                pass

        # Junction
        if CIGAR_op == "N":
                don = pos
                acc = pos + CIGAR_len
                if don > start and acc < end:
                        junctions[(don,acc)] = junctions.setdefault((don,acc), 0) + 1

        pos = pos + CIGAR_len

        return pos


def flip_read(s, samflag):
        if s == "NONE" or s == "SENSE":
                return 0
        if s == "ANTISENSE":
                return 1
        if s == "MATE1_SENSE":
                if int(samflag) & 64:
                        return 0
                if int(samflag) & 128:
                        return 1
        if s == "MATE2_SENSE":
                if int(samflag) & 64:
                        return 1
                if int(samflag) & 128:
                        return 0


def read_bam(f, c, s):

        _, start, end = parse_coordinates(c)

        # Initialize coverage array and junction dict
        a = {"+" : [0] * (end - start)}
        junctions = {"+": OrderedDict()}
        if s != "NONE":
                a["-"] = [0] * (end - start)
                junctions["-"] = OrderedDict()

        p = sp.Popen("samtools view %s %s " %(f, c), shell=True, stdout=sp.PIPE)
        for line in p.communicate()[0].decode('utf8').strip().split("\n"):

                if line == "":
                        continue

                line_sp = line.strip().split("\t")
                samflag, read_start, CIGAR = line_sp[1], int(line_sp[3]), line_sp[5]

                # Ignore reads with more exotic CIGAR operators
                if any(map(lambda x: x in CIGAR, ["H", "P", "X", "="])):
                        continue

                read_strand = ["+", "-"][flip_read(s, samflag) ^ bool(int(samflag) & 16)]
                if s == "NONE": read_strand = "+"

                CIGAR_lens = re.split("[MIDNS]", CIGAR)[:-1]
                CIGAR_ops = re.split("[0-9]+", CIGAR)[1:]

                pos = read_start

                for n, CIGAR_op in enumerate(CIGAR_ops):
                        CIGAR_len = int(CIGAR_lens[n])
                        pos = count_operator(CIGAR_op, CIGAR_len, pos, start, end, a[read_strand], junctions[read_strand])

        p.stdout.close()
        return a, junctions

def get_bam_path(index, path):
        if os.path.isabs(path):
                return path
        base_dir = os.path.dirname(index)
        return os.path.join(base_dir, path)

def read_bam_input(f, overlay, color, label):
        if f.endswith(".bam"):
                bn = f.strip().split("/")[-1].strip(".bam")
                yield bn, f, None, None, bn
                return
        with open(f) as openf:
                for line in openf:
                        line_sp = line.strip().split("\t")
                        bam = get_bam_path(f, line_sp[1])
                        overlay_level = line_sp[overlay-1] if overlay else None
                        color_level = line_sp[color-1] if color else None
                        label_text = line_sp[label-1] if label else None
                        yield line_sp[0], bam, overlay_level, color_level, label_text


def prepare_for_R(a, junctions, c, m):

        _, start, _ = parse_coordinates(args.coordinates)

        # Convert the array index to genomic coordinates
        x = list(i+start for i in range(len(a)))
        y = a

        # Arrays for R
        dons, accs, yd, ya, counts = [], [], [], [], []

        # Prepare arrays for junctions (which will be the arcs)
        for (don, acc), n in junctions.items():

                # Do not add junctions with less than defined coverage
                if n < m:
                        continue

                dons.append(don)
                accs.append(acc)
                counts.append(n)

                yd.append( a[ don - start -1 ])
                ya.append( a[ acc - start +1 ])

        return x, y, dons, accs, yd, ya, counts


def intersect_introns(data):
        data = sorted(data)
        it = iter(data)
        a, b = next(it)
        for c, d in it:
                if b > c:  # Use `if b > c` if you want (1,2), (2,3) not to be
                                # treated as intersection.
                        b = min(b, d)
                        a = max(a, c)
                else:
                        yield a, b
                        a, b = c, d
        yield a, b



def shrink_density(x, y, introns):
        new_x, new_y = [], []
        shift = 0
        start = 0
        # introns are already sorted by coordinates
        for a,b in introns:
                end = x.index(a)+1
                new_x += [int(i-shift) for i in x[start:end]]
                new_y += y[start:end]
                start = x.index(b)
                l = (b-a)
                shift += l-l**0.7
        new_x += [int(i-shift) for i in x[start:]]
        new_y += y[start:]
        return new_x, new_y

def shrink_junctions(dons, accs, introns):
        new_dons, new_accs = [0]*len(dons), [0]*len(accs)
        shift_acc = 0
        shift_don = 0
        s = set()
        junctions = list(zip(dons, accs))
        for a,b in introns:
                l = b - a
                shift_acc += l-int(l**0.7)
                for i, (don, acc) in enumerate(junctions):
                        if a >= don and b <= acc:
                                if (don,acc) not in s:
                                        new_dons[i] = don - shift_don
                                        new_accs[i] = acc - shift_acc
                                else:
                                        new_accs[i] = acc - shift_acc
                                s.add((don,acc))
                shift_don = shift_acc
        return new_dons, new_accs

def read_palette(f):
        palette = "#ff0000", "#00ff00", "#0000ff", "#000000"
        if f:
                with open(f) as openf:
                        palette = list(line.split("\t")[0].strip() for line in openf)
        return palette


def read_gtf(f, c):
        exons = OrderedDict()
        transcripts = OrderedDict()
        chr, start, end = parse_coordinates(c)
        end = end -1
        with open(f) as openf:
                for line in openf:
                        if line.startswith("#"):
                                continue
                        el_chr, _, el, el_start, el_end, _, strand, _, tags = line.strip().split("\t")
                        if el_chr != chr:
                                continue
                        d = dict(kv.strip().split(" ") for kv in tags.strip(";").split("; "))
                        transcript_id = d["transcript_id"]
                        el_start, el_end = int(el_start) -1, int(el_end)
                        strand = '"' + strand + '"'
                        if el == "transcript":
                                if (el_end > start and el_start < end):
                                        transcripts[transcript_id] = max(start, el_start), min(end, el_end), strand
                                continue
                        if el == "exon":
                                if (start < el_start < end or start < el_end < end):
                                        exons.setdefault(transcript_id, []).append((max(el_start, start), min(end, el_end), strand))

        return transcripts, exons


def make_introns(transcripts, exons, intersected_introns=None):
        new_transcripts = copy.deepcopy(transcripts)
        new_exons = copy.deepcopy(exons)
        introns = OrderedDict()
        if intersected_introns:
                for tx, (tx_start,tx_end,strand) in new_transcripts.items():
                        total_shift = 0
                        for a,b in intersected_introns:
                                l = b - a
                                shift = l - int(l**0.7)
                                total_shift += shift
                                for i, (exon_start,exon_end,strand) in enumerate(exons.get(tx,[])):
                                        new_exon_start, new_exon_end = new_exons[tx][i][:2]
                                        if a < exon_start:
                                                if b > exon_end:
                                                        if i ==  len(exons[tx])-1:
                                                                total_shift = total_shift - shift + (exon_start - a)*(1-int(l**-0.3))
                                                        shift = (exon_start - a)*(1-int(l**-0.3))
                                                        new_exon_end = new_exons[tx][i][1] - shift
                                                new_exon_start = new_exons[tx][i][0] - shift
                                        if b <= exon_end:
                                                new_exon_end = new_exons[tx][i][1] - shift
                                        new_exons[tx][i] = (new_exon_start,new_exon_end,strand)
                        tx_start = min(tx_start, sorted(new_exons.get(tx, [[sys.maxsize]]))[0][0])
                        new_transcripts[tx] = (tx_start, tx_end - total_shift, strand)

        for tx, (tx_start,tx_end,strand) in new_transcripts.items():
                intron_start = tx_start
                ex_end = 0
                for ex_start, ex_end, strand in sorted(new_exons.get(tx, [])):
                        intron_end = ex_start
                        if tx_start < ex_start:
                                introns.setdefault(tx, []).append((intron_start, intron_end, strand))
                        intron_start = ex_end
                if tx_end > ex_end:
                        introns.setdefault(tx, []).append((intron_start, tx_end, strand))
        d = {'transcripts': new_transcripts, 'exons': new_exons, 'introns': introns}
        return d


def gtf_for_ggplot(annotation, start, end, arrow_bins):
        arrow_space = int((end - start)/arrow_bins)
        s = """

        # data table with exons
        ann_list = list(
                "exons" = data.table(),
                "introns" = data.table()
        )
        """

        if annotation["exons"]:

                s += """
                ann_list[['exons']] = data.table(
                        tx = rep(c(%(tx_exons)s), c(%(n_exons)s)),
                        start = c(%(exon_start)s),
                        end = c(%(exon_end)s),
                        strand = c(%(strand)s)
                )
                """ %({
                "tx_exons": ",".join(annotation["exons"].keys()),
                "n_exons": ",".join(map(str, map(len, annotation["exons"].values()))),
                "exon_start" : ",".join(map(str, (v[0] for vs in annotation["exons"].values() for v in vs))),
                "exon_end" : ",".join(map(str, (v[1] for vs in annotation["exons"].values() for v in vs))),
                "strand" : ",".join(map(str, (v[2] for vs in annotation["exons"].values() for v in vs))),
                })

        if annotation["introns"]:

                s += """
                ann_list[['introns']] = data.table(
                        tx = rep(c(%(tx_introns)s), c(%(n_introns)s)),
                        start = c(%(intron_start)s),
                        end = c(%(intron_end)s),
                        strand = c(%(strand)s)
                )
                # Create data table for strand arrows
                txarrows = data.table()
                introns = ann_list[['introns']]
                # Add right-pointing arrows for plus strand
                if ("+" %%in%% introns$strand) {
                        txarrows = rbind(
                                txarrows,
                                introns[strand=="+" & end-start>5, list(
                                        seq(start+4,end,by=%(arrow_space)s)-1,
                                        seq(start+4,end,by=%(arrow_space)s)
                                        ), by=.(tx,start,end)
                                ]
                        )
                }
                # Add left-pointing arrows for minus strand
                if ("-" %%in%% introns$strand) {
                        txarrows = rbind(
                                txarrows,
                                introns[strand=="-" & end-start>5, list(
                                        seq(start,max(start+1, end-4), by=%(arrow_space)s),
                                        seq(start,max(start+1, end-4), by=%(arrow_space)s)-1
                                        ), by=.(tx,start,end)
                                ]
                        )
                }
                """ %({
                        "tx_introns": ",".join(annotation["introns"].keys()),
                        "n_introns": ",".join(map(str, map(len, annotation["introns"].values()))),
                        "intron_start" : ",".join(map(str, (v[0] for vs in annotation["introns"].values() for v in vs))),
                        "intron_end" : ",".join(map(str, (v[1] for vs in annotation["introns"].values() for v in vs))),
                        "strand" : ",".join(map(str, (v[2] for vs in annotation["introns"].values() for v in vs))),
                        "arrow_space" : arrow_space,
                })

        s += """

        gtfp = ggplot()
        if (length(ann_list[['introns']]) > 0) {
                gtfp = gtfp + geom_segment(data=ann_list[['introns']], aes(x=start, xend=end, y=tx, yend=tx), size=0.3)
                gtfp = gtfp + geom_segment(data=txarrows, aes(x=V1,xend=V2,y=tx,yend=tx), arrow=arrow(length=unit(0.02,"npc")))
        }
        if (length(ann_list[['exons']]) > 0) {
                gtfp = gtfp + geom_segment(data=ann_list[['exons']], aes(x=start, xend=end, y=tx, yend=tx), size=5, alpha=1)
        }
        gtfp = gtfp + scale_y_discrete(expand=c(0,0.5))
        gtfp = gtfp + scale_x_continuous(expand=c(0,0.25), limits = c(%s,%s))
        gtfp = gtfp + labs(y=NULL)
        gtfp = gtfp + theme(axis.line = element_blank(), axis.text.x = element_blank(), axis.ticks = element_blank())
        """ %(start, end)

        return s


def setup_R_script(h, w, b, label_dict):
        s = """
        library(ggplot2)
        library(grid)
        library(gridExtra)
        library(data.table)
        library(gtable)

        scale_lwd = function(r) {
                lmin = 0.1
                lmax = 4
                return( r*(lmax-lmin)+lmin )
        }

        base_size = %(b)s
        height = ( %(h)s + base_size*0.352777778/67 ) * 1.02
        width = %(w)s
        theme_set(theme_bw(base_size=base_size))
        theme_update(
                plot.margin = unit(c(15,15,15,15), "pt"),
                panel.grid = element_blank(),
                panel.border = element_blank(),
                axis.line = element_line(size=0.5),
                axis.title.x = element_blank(),
                axis.title.y = element_text(angle=0, vjust=0.5)
        )

        labels = list(%(labels)s)

        density_list = list()
        junction_list = list()

        """ %({
                'h': h,
                'w': w,
                'b': b,
                'labels': ",".join(('"%s"="%s"' %(id,lab) for id,lab in label_dict.items())),
        })
        return s

def median(lst):
    quotient, remainder = divmod(len(lst), 2)
    if remainder:
        return sorted(lst)[quotient]
    return sum(sorted(lst)[quotient - 1:quotient + 1]) / 2.

def mean(lst):
        return sum(lst)/len(lst)


def make_R_lists(id_list, d, overlay_dict, aggr, intersected_introns):
        s = ""
        aggr_f = {
                "mean": mean,
                "median": median,
        }
        id_list = id_list if not overlay_dict else overlay_dict.keys()
        # Iterate over ids to get bam signal and junctions
        for k in id_list:
                x, y, dons, accs, yd, ya, counts = [], [], [], [], [], [], []
                if not overlay_dict:
                        x, y, dons, accs, yd, ya, counts = d[k]
                        if intersected_introns:
                                x, y = shrink_density(x, y, intersected_introns)
                                dons, accs = shrink_junctions(dons, accs, intersected_introns)
                else:
                        for id in overlay_dict[k]:
                                xid, yid, donsid, accsid, ydid, yaid, countsid = d[id]
                                if intersected_introns:
                                        xid, yid = shrink_density(xid, yid, intersected_introns)
                                        donsid, accsid = shrink_junctions(donsid, accsid, intersected_introns)
                                x += xid
                                y += yid
                                dons += donsid
                                accs += accsid
                                yd += ydid
                                ya += yaid
                                counts += countsid
                        if aggr and "_j" not in aggr:
                                x = d[overlay_dict[k][0]][0]
                                y = list(map(aggr_f[aggr], zip(*(d[id][1] for id in overlay_dict[k]))))
                                if intersected_introns:
                                        x, y = shrink_density(x, y, intersected_introns)
                        #dons, accs, yd, ya, counts = [], [], [], [], []
                s += """
                density_list[["%(id)s"]] = data.frame(x=c(%(x)s), y=c(%(y)s))
                junction_list[["%(id)s"]] = data.frame(x=c(%(dons)s), xend=c(%(accs)s), y=c(%(yd)s), yend=c(%(ya)s), count=c(%(counts)s))
                """ %({
                        "id": k,
                        'x' : ",".join(map(str, x)),
                        'y' : ",".join(map(str, y)),
                        'dons' : ",".join(map(str, dons)),
                        'accs' : ",".join(map(str, accs)),
                        'yd' : ",".join(map(str, yd)),
                        'ya' : ",".join(map(str, ya)),
                        'counts' : ",".join(map(str, counts)),
                })
        return s


def plot(R_script):
        p = sp.Popen("R --vanilla --slave", shell=True, stdin=sp.PIPE)
        p.communicate(input=R_script.encode())
        p.stdin.close()
        p.wait()
        return


def colorize(d, p, color_factor):
        levels = sorted(set(d.values()))
        n = len(levels)
        if n > len(p):
                p = (p*n)[:n]
        if color_factor:
                s = "color_list = list(%s)\n" %( ",".join('%s="%s"' %(k, p[levels.index(v)]) for k,v in d.items()) )
        else:
                s = "color_list = list(%s)\n" %( ",".join('%s="%s"' %(k, "grey") for k,v in d.items()) )
        return s



if __name__ == "__main__":

        strand_dict = {"plus": "+", "minus": "-"}

        parser = define_options()
        if len(sys.argv)==1:
            parser.print_help()
            sys.exit(1)
        args = parser.parse_args()

#       args.coordinates = "chrX:9609491-9612406"
#       args.coordinates = "chrX:9609491-9610000"
#       args.bam = "/nfs/no_backup/rg/epalumbo/projects/tg/work/8b/8b0ac8705f37fd772a06ab7db89f6b/2A_m4_n10_toGenome.bam"

        if args.aggr and not args.overlay:
                print("ERROR: Cannot apply aggregate function if overlay is not selected.")
                exit(1)

        palette = read_palette(args.palette)

        bam_dict, overlay_dict, color_dict, id_list, label_dict = {"+":OrderedDict()}, OrderedDict(), OrderedDict(), [], OrderedDict()
        if args.strand != "NONE": bam_dict["-"] = OrderedDict()
        if args.junctions_bed != "": junctions_list = []

        for id, bam, overlay_level, color_level, label_text in read_bam_input(args.bam, args.overlay, args.color_factor, args.labels):
                if not os.path.isfile(bam):
                        continue
                id_list.append(id)
                label_dict[id] = label_text
                a, junctions = read_bam(bam, args.coordinates, args.strand)
                if a.keys() == ["+"] and all(map(lambda x: x==0, list(a.values()[0]))):
                        print("ERROR: No reads in the specified area.")
                        exit(1)
                for strand in a:
                        # Store junction information                 
                        if args.junctions_bed: 
                                for k,v in zip(junctions[strand].keys(), junctions[strand].values()):
                                        if v > args.min_coverage:
                                                junctions_list.append('\t'.join([args.coordinates.split(':')[0], str(k[0]), str(k[1]), id, str(v), strand]))                     
                        bam_dict[strand][id] = prepare_for_R(a[strand], junctions[strand], args.coordinates, args.min_coverage)
                if color_level is None:
                        color_dict.setdefault(id, id)
                if overlay_level is not None:
                        overlay_dict.setdefault(overlay_level, []).append(id)
                        label_dict[overlay_level] = overlay_level
                        color_dict.setdefault(overlay_level, overlay_level)
                if overlay_level is None:
                        color_dict.setdefault(id, color_level)

        # No bam files
        if not bam_dict["+"]:
                print("ERROR: No available bam files.")
                exit(1)

        # Write junctions to BED
        if args.junctions_bed: 
                if not args.junctions_bed.endswith('.bed'):
                        args.junctions_bed = args.junctions_bed + '.bed'
                jbed = open(args.junctions_bed, 'w')
                jbed.write('\n'.join(sorted(junctions_list)))
                jbed.close()

        if args.gtf:
                transcripts, exons = read_gtf(args.gtf, args.coordinates)

        if args.out_format not in ('pdf', 'png', 'svg', 'tiff', 'jpeg'):
                print("ERROR: Provided output format '%s' is not available. Please select among 'pdf', 'png', 'svg', 'tiff' or 'jpeg'" % args.out_format)
                exit(1)

        # Iterate for plus and minus strand
        for strand in bam_dict:

                # Output file name (allow tiff/tif and jpeg/jpg extensions)
                if args.out_prefix.endswith(('.pdf', '.png', '.svg', '.tiff', '.tif', '.jpeg', '.jpg')):
                        out_split = os.path.splitext(args.out_prefix)
                        if (args.out_format == out_split[1][1:] or
                        args.out_format == 'tiff' and out_split[1] in ('.tiff','.tif') or
                        args.out_format == 'jpeg' and out_split[1] in ('.jpeg','.jpg')):
                                args.out_prefix = out_split[0]
                                out_suffix = out_split[1][1:]
                        else:
                                out_suffix = args.out_format
                else:
                        out_suffix = args.out_format
                out_prefix = args.out_prefix + "_" + strand
                if args.strand == "NONE":
                        out_prefix = args.out_prefix
                else:
                        if args.out_strand != "both" and strand != strand_dict[args.out_strand]:
                                continue

                # Find set of junctions to perform shrink
                intersected_introns = None
                if args.shrink:
                        introns = (v for vs in bam_dict[strand].values() for v in zip(vs[2], vs[3]))
                        intersected_introns = list(intersect_introns(introns))


                # *** PLOT *** Define plot height
                bam_height = args.height * len(id_list)
                if args.overlay:
                        bam_height = args.height * len(overlay_dict)
                if args.gtf:
                        bam_height += args.ann_height

                # *** PLOT *** Start R script by loading libraries, initializing variables, etc...
                R_script = setup_R_script(bam_height, args.width, args.base_size, label_dict)

                R_script += colorize(color_dict, palette, args.color_factor)

                # *** PLOT *** Prepare annotation plot only for the first bam file
                arrow_bins = 50
                if args.gtf:
                        # Make introns from annotation (they are shrunk if required)
                        annotation = make_introns(transcripts, exons, intersected_introns)
                        x = list(bam_dict[strand].values())[0][0]
                        if args.shrink:
                                x, _ = shrink_density(x, x, intersected_introns)
                        R_script += gtf_for_ggplot(annotation, x[0], x[-1], arrow_bins)

                R_script += make_R_lists(id_list, bam_dict[strand], overlay_dict, args.aggr, intersected_introns)

                R_script += """

                pdf(NULL) # just to remove the blank pdf produced by ggplotGrob

                if(packageVersion('ggplot2') >= '3.0.0'){  # fix problems with ggplot2 vs >3.0.0
                        vs = 1
                } else {
                        vs = 0
                }

                density_grobs = list();

                      for (bam_index in 1:length(density_list)) {

                        id = names(density_list)[bam_index]
                        d = data.table(density_list[[id]])
                        junctions = data.table(junction_list[[id]])

                        maxheight = max(d[['y']])

                        # Density plot
                        gp = ggplot(d) + geom_bar(aes(x, y), width=1, position='identity', stat='identity', fill=color_list[[id]], alpha=%(alpha)s)
                        gp = gp + labs(y=labels[[id]])
                        gp = gp + scale_x_continuous(expand=c(0,0.2))

                        if(packageVersion('ggplot2') >= '3.0.0'){ # fix problems with ggplot2 vs >3.0.0
                                gp = gp + scale_y_continuous(breaks=ggplot_build(gp)$layout$panel_params[[1]]$y.major_source)
                        } else {
                                gp = gp + scale_y_continuous(breaks=ggplot_build(gp)$layout$panel_ranges[[1]]$y.major_source)
                        }

                        # Aggregate junction counts
                        row_i = c()
                        if (nrow(junctions) >0 ) {

                                junctions$jlabel = as.character(junctions$count)
                                junctions = setNames(junctions[,.(max(y), max(yend),round(mean(count)),paste(jlabel,collapse=",")), keyby=.(x,xend)], names(junctions))
                                if ("%(args.aggr)s" != "") {
                                        junctions = setNames(junctions[,.(max(y), max(yend),round(%(args.aggr)s(count)),round(%(args.aggr)s(count))), keyby=.(x,xend)], names(junctions))
                                }
                                # The number of rows (unique junctions per bam) has to be calculated after aggregation
                                row_i = 1:nrow(junctions)
                        }


                        for (i in row_i) {

                                j_tot_counts = sum(junctions[['count']])

                                j = as.numeric(junctions[i,1:5])

                                if ("%(args.aggr)s" != "") {
                                        j[3] = as.numeric(d[x==j[1]-1,y])
                                        j[4] = as.numeric(d[x==j[2]+1,y])
                                }

                                # Find intron midpoint
                                xmid = round(mean(j[1:2]), 1)
                                ymid = max(j[3:4]) * 1.2

                                # Thickness of the arch
                                lwd = scale_lwd(j[5]/j_tot_counts)

                                curve_par = gpar(lwd=lwd, col=color_list[[id]])

                                # Arc grobs

                                # Choose position of the arch (top or bottom)
                                nss = i
                                if (nss%%%%2 == 0) {  #bottom
                                        ymid = -0.3 * maxheight
                                        # Draw the arcs
                                        # Left
                                        curve = xsplineGrob(x=c(0, 0, 1, 1), y=c(1, 0, 0, 0), shape=1, gp=curve_par)
                                        gp = gp + annotation_custom(grob = curve, j[1], xmid, 0, ymid)
                                        # Right
                                        curve = xsplineGrob(x=c(1, 1, 0, 0), y=c(1, 0, 0, 0), shape=1, gp=curve_par)
                                        gp = gp + annotation_custom(grob = curve, xmid, j[2], 0, ymid)
                                }

                                if (nss%%%%2 != 0) {  #top
                                        # Draw the arcs
                                        # Left
                                        curve = xsplineGrob(x=c(0, 0, 1, 1), y=c(0, 1, 1, 1), shape=1, gp=curve_par)
                                        gp = gp + annotation_custom(grob = curve, j[1], xmid, j[3], ymid)
                                        # Right
                                        curve = xsplineGrob(x=c(1, 1, 0, 0), y=c(0, 1, 1, 1), shape=1, gp=curve_par)
                                        gp = gp + annotation_custom(grob = curve, xmid, j[2], j[4], ymid)
                                }

                                # Add junction labels
                                gp = gp + annotate("label", x = xmid, y = ymid, label = as.character(junctions[i,6]),
                                        vjust=0.5, hjust=0.5, label.padding=unit(0.01, "lines"),
                                        label.size=NA, size=(base_size*0.352777778)*0.6
                                )


                #               gp = gp + annotation_custom(grob = rectGrob(x=0, y=0, gp=gpar(col="red"), just=c("left","bottom")), xmid, j[2], j[4], ymid)
                #               gp = gp + annotation_custom(grob = rectGrob(x=0, y=0, gp=gpar(col="green"), just=c("left","bottom")), j[1], xmid, j[3], ymid)


                        }

                        gpGrob = ggplotGrob(gp);
                        gpGrob$layout$clip[gpGrob$layout$name=="panel"] <- "off"
                        if (bam_index == 1) {
                                maxWidth = gpGrob$widths[2+vs] + gpGrob$widths[3+vs];    # fix problems ggplot2 vs
                                maxYtextWidth = gpGrob$widths[3+vs];                     # fix problems ggplot2 vs
                                # Extract x axis grob (trim=F --> keep empty cells)
                                xaxisGrob <- gtable_filter(gpGrob, "axis-b", trim=F)
                                xaxisGrob$heights[8+vs] = gpGrob$heights[1]              # fix problems ggplot2 vs
                                x.axis.height = gpGrob$heights[7+vs] + gpGrob$heights[1] # fix problems ggplot2 vs
                        }


                        # Remove x axis from all density plots
                        kept_names = gpGrob$layout$name[gpGrob$layout$name != "axis-b"]
                        gpGrob <- gtable_filter(gpGrob, paste(kept_names, sep="", collapse="|"), trim=F)

                        # Find max width of y text and y label and max width of y text
                        maxWidth = grid::unit.pmax(maxWidth, gpGrob$widths[2+vs] + gpGrob$widths[3+vs]); # fix problems ggplot2 vs
                        maxYtextWidth = grid::unit.pmax(maxYtextWidth, gpGrob$widths[3+vs]); # fix problems ggplot2 vs
                        density_grobs[[id]] = gpGrob;
                }

                # Add x axis grob after density grobs BEFORE annotation grob
                density_grobs[["xaxis"]] = xaxisGrob

                # Annotation grob
                if (%(args.gtf)s == 1) {
                        gtfGrob = ggplotGrob(gtfp);
                        maxWidth = grid::unit.pmax(maxWidth, gtfGrob$widths[2+vs] + gtfGrob$widths[3+vs]); # fix problems ggplot2 vs
                        density_grobs[['gtf']] = gtfGrob;
                }

                # Reassign grob widths to align the plots
                for (id in names(density_grobs)) {
                        density_grobs[[id]]$widths[1] <- density_grobs[[id]]$widths[1] + maxWidth - (density_grobs[[id]]$widths[2+vs] + maxYtextWidth); # fix problems ggplot2 vs
                        density_grobs[[id]]$widths[3+vs] <- maxYtextWidth # fix problems ggplot2 vs
                }

                # Heights for density, x axis and annotation
                heights = unit.c(
                        unit(rep(%(signal_height)s, length(density_list)), "in"),
                        x.axis.height,
                        unit(%(ann_height)s*%(args.gtf)s, "in")
                        )

                # Arrange grobs
                argrobs = arrangeGrob(
                        grobs=density_grobs,
                        ncol=1,
                        heights = heights,
                );

                # Save plot to file in the requested format
                if ("%(out_format)s" == "tiff"){
                        # TIFF images will be lzw-compressed
                        ggsave("%(out)s", plot = argrobs, device = "tiff", width = width, height = height, units = "in", dpi = %(out_resolution)s, compression = "lzw")
                } else {
                        ggsave("%(out)s", plot = argrobs, device = "%(out_format)s", width = width, height = height, units = "in", dpi = %(out_resolution)s)
                }

                dev.log = dev.off()

                """ %({
                        "out": "%s.%s" % (out_prefix, out_suffix),
                        "out_format": args.out_format,
                        "out_resolution": args.out_resolution,
                        "args.gtf": float(bool(args.gtf)),
                        "args.aggr": args.aggr.rstrip("_j"),
                        "signal_height": args.height,
                        "ann_height": args.ann_height,
                        "alpha": args.alpha,
                        })
                if os.getenv('GGSASHIMI_DEBUG') is not None:
                        with open("R_script", 'w') as r:
                                r.write(R_script)
                else:
                        plot(R_script)
        exit()