def getRawCoverageLargeMode(self, chrom, start, end, largeMode=10):
"""Retrieve an array with the genome coverage."""
out = np.zeros(end-start)
infile = BigWigFile(self.path)
wigs = infile.fetch(chrom, start, end)
for wig in wigs:
out[wig[0]-start:wig[1]-start] = wig[2]
infile.close()
return out[::largeMode]
def bigwig_fetcher(bigwig, ichr, istart, iend): bw=BigWigFile(bigwig) scores = [] with BigWigFile(bigwig) as bw: for i in bw.fetch(chrom=ichr,start=istart,stop=iend): scores.append(i.score) return scores bw.wWigIO.close()
def coverage_from_bigwig(self, bigwig_file, stepsize=100):
"""Return list of arrays describing the coverage of each genomicRegions from <bigwig_file>.
*Keyword arguments:*
- bigwig_file -- path to bigwig file
- stepsize -- used stepsize
*Output:*
Class variable <coverage>: a list where the elements correspond to the GenomicRegion. The list elements give
the number of reads falling into the GenomicRegion.
"""
if platform == "darwin" or "http" in bigwig_file:
self.coverage = []
# mp_input = []
for gr in self.genomicRegions:
# print(gr)
steps = int(abs(gr.final-gr.initial)/stepsize)
cmd = ["bigWigSummary",bigwig_file,gr.chrom,str(gr.initial-stepsize),str(gr.final-stepsize),str(steps)]
# print(" ".join(cmd))
try:
output = subprocess.check_output(cmd, shell=False, stderr=subprocess.STDOUT)
# print(output)
ds = [0 if "n/a" in x else float(x) for x in output.strip().split()]
self.coverage.append( np.array(ds) )
except:
continue
### Linux platform
else:
# print("\tUsing ngslib on linux system...")
from ngslib import BigWigFile
self.coverage = []
bwf = BigWigFile(bigwig_file)
for gr in self.genomicRegions:
depth = bwf.pileup(gr.chrom, max(0,int(gr.initial-stepsize/2)),
max(1,int(gr.final+stepsize/2)))
ds = [depth[d] for d in range(0, gr.final-gr.initial, stepsize)]
self.coverage.append( np.array(ds) )
bwf.close()
def bigwig_loader(bigwig, chromInfo): print >> sys.stderr, "Loading the BigWig file in RAM memory ...", bw=BigWigFile(bigwig) for row in csv.reader(open(chromInfo), delimiter = '\t'): chr = row[0] chr_size = int(row[1]) for i in bw.fetch(chrom=chr,start=0,stop=chr_size): phylop_bigwig[chr].append(i.score) bw.wWigIO.close() print >> sys.stderr, "OK",
def __init__(self, src_file, title='', version=None):
# only necessary to import ngslib if instance of BigWigDatasource is created
# This should not run on OS X machines
from ngslib import BigWigFile
super(BigWigDatasource, self).__init__(src_file, title=title, version=version)
self.output_headers = [title + '_score']
self.bigwig_fh = BigWigFile(src_file)
self.has_chr = True if self.bigwig_fh.chroms[0].startswith('chr') else False
class BigWigDatasource(Datasource):
"""
A datasource derived from a BigWig file. For variants spanning a genomic range (i.e. non SNVs),
the median of values from the BigWig are returned.
"""
def __init__(self, src_file, title='', version=None):
# only necessary to import ngslib if instance of BigWigDatasource is created
# This should not run on OS X machines
from ngslib import BigWigFile
super(BigWigDatasource, self).__init__(src_file, title=title, version=version)
self.output_headers = [title + '_score']
self.bigwig_fh = BigWigFile(src_file)
self.has_chr = True if self.bigwig_fh.chroms[0].startswith('chr') else False
def annotate_mutation(self, mutation):
if self.has_chr and not mutation.chr.startswith('chr'):
chrn = 'chr' + mutation.chr
else:
chrn = mutation.chr
variant_start, variant_end = int(mutation.start) - 1, int(mutation.end) #start - 1 because bigwig format is zero-based coords
scores = [r[2] for r in self.bigwig_fh.fetch(chrom=chrn, start=variant_start, stop=variant_end)]
if not scores:
final_score = None
elif len(scores) == 1:
final_score = scores[0]
else:
final_score = np.median(scores)
mutation.createAnnotation(self.output_headers[0], final_score, annotationSource=self.title)
return mutation
def close(self):
self.bigwig_fh.close()
def phastCons46way_score(self, stepsize=100):
"""Load the phastCons46way bigwig files to fetch the scores as coverage.
*Keyword arguments:*
- stepsize -- used stepsize
"""
self.coverage = []
phastCons46way_dir = "/data/phastCons46way/"
for gr in self.genomicRegions:
bwf = BigWigFile(os.path.join(phastCons46way_dir, gr.chrom+".phastCons46way.bw"))
depth = bwf.pileup(gr.chrom, gr.initial-stepsize/2, gr.final+stepsize/2)
ds = []
for i in range(0, gr.final-gr.initial):
d = [ depth[j] for j in range(i,i+stepsize) ]
ds.append(sum(d)/len(d))
if gr.orientation == "-":
self.coverage.append( np.array(list(reversed(ds))) )
else:
self.coverage.append( np.array(ds) )
bwf.close()
def bigwig_mean(bigwig, chr, start, end): bw=BigWigFile(bigwig) score_sum = 0 mean_score = 0 with BigWigFile(bigwig) as bw: for i in bw.fetch(chrom=chr,start=start,stop=end): score_sum += i.score if (end-start) != 0: mean_score = score_sum/(end-start) else: mean_score = 0 return mean_score bw.wWigIO.close()
def coverage_from_bigwig(self, bigwig_file, stepsize=100):
"""Return list of arrays describing the coverage of each genomicRegions from <bigwig_file>.
*Keyword arguments:*
- bigwig_file -- path to bigwig file
- stepsize -- used stepsize
*Output:*
Class variable <coverage>: a list where the elements correspond to the GenomicRegion. The list elements give
the number of reads falling into the GenomicRegion.
"""
try:
from ngslib import BigWigFile
self.coverage = []
bwf = BigWigFile(bigwig_file)
for gr in self.genomicRegions:
depth = bwf.pileup(gr.chrom,
max(0, int(gr.initial - stepsize / 2)),
max(1, int(gr.final + stepsize / 2)))
ds = [
depth[d] for d in range(0, gr.final - gr.initial, stepsize)
]
self.coverage.append(np.array(ds))
bwf.close()
except ImportError, e:
import pyBigWig
self.coverage = []
bwf = pyBigWig.open(bigwig_file)
for gr in self.genomicRegions:
steps = int(len(gr) / stepsize)
ds = bwf.stats(gr.chrom,
gr.initial,
gr.final,
type="mean",
nBins=steps)
ds = [x if x else 0 for x in ds]
self.coverage.append(np.array(ds))
bwf.close()
def __init__(self, file_name):
"""
Initializes GenomicSignal.
"""
self.file_name = file_name
self.bam = None
self.bw = None
self.sg_coefs = None
self.is_bam = False
self.is_bw = False
if(self.file_name.split(".")[-1].upper() == "BAM"):
self.is_bam = True
self.bam = Samfile(file_name,"rb")
elif(self.file_name.split(".")[-1].upper() == "BW" or self.file_name.split(".")[-1].upper() == "BIGWIG"):
self.is_bw = True
self.bw = BigWigFile(file_name)
else: pass # TODO ERROR
def coverage_from_bigwig(self, bigwig_file, stepsize=100):
"""Return list of arrays describing the coverage of each genomicRegions from <bigwig_file>.
*Keyword arguments:*
- bigwig_file -- path to bigwig file
- stepsize -- used stepsize
*Output:*
Class variable <coverage>: a list where the elements correspond to the GenomicRegion. The list elements give
the number of reads falling into the GenomicRegion.
"""
try:
from ngslib import BigWigFile
self.coverage = []
bwf = BigWigFile(bigwig_file)
for gr in self.genomicRegions:
depth = bwf.pileup(gr.chrom, max(0, int(gr.initial - stepsize / 2)),
max(1, int(gr.final + stepsize / 2)))
ds = [depth[d] for d in range(0, gr.final - gr.initial, stepsize)]
self.coverage.append(np.array(ds))
bwf.close()
except ImportError, e:
import pyBigWig
self.coverage = []
bwf = pyBigWig.open(bigwig_file)
for gr in self.genomicRegions:
steps = int(len(gr) / stepsize)
ds = bwf.stats(gr.chrom, gr.initial, gr.final, type="mean", nBins=steps)
ds = [ x if x else 0 for x in ds ]
self.coverage.append( np.array(ds) )
bwf.close()
def getChromSizesNGSLIB(self):
infile = BigWigFile(self.path)
out = infile.chromSizes()
out = dict(zip(out[0], out[1]))
infile.close()
return out
class GenomicSignal:
"""
Represents a genomic signal. It should be used to fetch normalized and slope
signals from a bam or bw file.
Usage:
1. Initialize class.
2. Call load_sg_coefs once.
3. Call get_signal as many times as needed.
Authors: Eduardo G. Gusmao.
Methods:
load_sg_coefs(self, slope_window_size):
Loads Savitzky-Golay coefficients into self.sg_coefs based on a slope_window_size.
get_signal(self, ref, start, end, ext, initial_clip = 1000, per_norm = 98, per_slope = 98)
Gets the signal associated with self.bam or self.bw based on start, end and ext.
initial_clip, per_norm and per_slope are used as normalization factors during the normalization
and slope evaluation procedures.
hon_norm(self, sequence, mean, std):
Normalizes a sequence according to hon's criterion using mean and std.
This represents a between-dataset normalization.
boyle_norm(self, sequence):
Normalizes a sequence according to Boyle's criterion.
This represents a within-dataset normalization.
savitzky_golay_coefficients(self, window_size, order, deriv):
Evaluate the Savitzky-Golay coefficients in order to evaluate the slope of the signal.
It uses a window_size (of the interpolation), order (of the polynomial), deriv (derivative needed).
slope(self, sequence, sg_coefs):
Evaluates the slope of sequence given the sg_coefs loaded.
"""
def __init__(self, file_name):
"""
Initializes GenomicSignal.
"""
self.file_name = file_name
self.bam = None
self.bw = None
self.sg_coefs = None
self.is_bam = False
self.is_bw = False
if(self.file_name.split(".")[-1].upper() == "BAM"):
self.is_bam = True
self.bam = Samfile(file_name,"rb")
elif(self.file_name.split(".")[-1].upper() == "BW" or self.file_name.split(".")[-1].upper() == "BIGWIG"):
self.is_bw = True
self.bw = BigWigFile(file_name)
else: pass # TODO ERROR
def load_sg_coefs(self, slope_window_size):
"""
Loads Savitzky-Golay coefficients into self.sg_coefs based on a slope_window_size.
Keyword arguments:
slope_window_size -- Window size of Savitzky-Golay coefficients.
Return:
None -- It updates self.sg_coefs.
"""
self.sg_coefs = self.savitzky_golay_coefficients(slope_window_size, 2, 1)
def get_tag_count(self, ref, start, end, ext, initial_clip = 1000, ext_both_directions=False):
"""
Gets the tag count associated with self.bam based on start, end and ext.
Keyword arguments:
ref -- Chromosome name.
start -- Initial genomic coordinate of signal.
end -- Final genomic coordinate of signal.
ext -- Fragment extention. Eg. 1 for DNase and 200 for histone modifications.
initial_clip -- Signal will be initially clipped at this level to avoid outliers.
Return:
tag_count -- Total signal.
"""
# Fetch raw signal
pileup_region = PileupRegion(start,end,ext)
if(self.is_bam):
if(ps_version == "0.7.5"):
self.bam.fetch(reference=ref, start=start, end=end, callback = pileup_region)
else:
iter = self.bam.fetch(reference=ref, start=start, end=end)
if(not ext_both_directions):
for alignment in iter: pileup_region.__call__(alignment)
else:
for alignment in iter: pileup_region.__call2__(alignment)
raw_signal = array([min(e,initial_clip) for e in pileup_region.vector])
elif(self.is_bw):
signal = self.bw.pileup(ref, start, end)
raw_signal = array([min(e,initial_clip) for e in signal])
# Std-based clipping
mean = raw_signal.mean()
std = raw_signal.std()
clip_signal = [min(e, mean + (10 * std)) for e in raw_signal]
# Tag count
try: tag_count = sum(clip_signal)
except Exception: tag_count = 0
return tag_count
def get_signal(self, ref, start, end, ext, initial_clip = 1000, per_norm = 99.5, per_slope = 98,
bias_table = None, genome_file_name = None, ext_both_directions=False, print_wig = None):
"""
Gets the signal associated with self.bam based on start, end and ext.
initial_clip, per_norm and per_slope are used as normalization factors during the normalization
and slope evaluation procedures.
Keyword arguments:
ref -- Chromosome name.
start -- Initial genomic coordinate of signal.
end -- Final genomic coordinate of signal.
ext -- Fragment extention. Eg. 1 for DNase and 200 for histone modifications.
initial_clip -- Signal will be initially clipped at this level to avoid outliers.
per_norm -- Percentile value for 'hon_norm' function of the normalized signal.
per_slope -- Percentile value for 'hon_norm' function of the slope signal.
bias_table -- Bias table to perform bias correction.
Return:
hon_signal -- Normalized signal.
slopehon_signal -- Slope signal.
"""
# Fetch raw signal
pileup_region = PileupRegion(start,end,ext)
if(self.is_bam):
if(ps_version == "0.7.5"):
self.bam.fetch(reference=ref, start=start, end=end, callback = pileup_region)
else:
iter = self.bam.fetch(reference=ref, start=start, end=end)
if(not ext_both_directions):
for alignment in iter: pileup_region.__call__(alignment)
else:
for alignment in iter: pileup_region.__call2__(alignment)
raw_signal = array([min(e,initial_clip) for e in pileup_region.vector])
elif(self.is_bw):
signal = self.bw.pileup(ref, start, end)
raw_signal = array([min(e,initial_clip) for e in signal])
# Std-based clipping
mean = raw_signal.mean()
std = raw_signal.std()
clip_signal = [min(e, mean + (10 * std)) for e in raw_signal]
# Bias correction
bias_corrected_signal = self.bias_correction(clip_signal, bias_table, genome_file_name, ref, start, end)
# Boyle normalization (within-dataset normalization)
boyle_signal = array(self.boyle_norm(bias_corrected_signal))
# Hon normalization (between-dataset normalization)
perc = scoreatpercentile(boyle_signal, per_norm)
std = boyle_signal.std()
hon_signal = self.hon_norm(boyle_signal, perc, std)
# Slope signal
slope_signal = self.slope(hon_signal, self.sg_coefs)
# Hon normalization on slope signal (between-dataset slope smoothing)
abs_seq = array([abs(e) for e in slope_signal])
perc = scoreatpercentile(abs_seq, per_slope)
std = abs_seq.std()
slopehon_signal = self.hon_norm(slope_signal, perc, std)
# Writing signal
if(print_wig):
signal_file = open(print_wig+"signal.wig","a")
norm_file = open(print_wig+"norm.wig","a")
slope_file = open(print_wig+"slope.wig","a")
signal_file.write("fixedStep chrom="+ref+" start="+str(start+1)+" step=1\n"+"\n".join([str(e) for e in clip_signal])+"\n")
norm_file.write("fixedStep chrom="+ref+" start="+str(start+1)+" step=1\n"+"\n".join([str(e) for e in hon_signal])+"\n")
slope_file.write("fixedStep chrom="+ref+" start="+str(start+1)+" step=1\n"+"\n".join([str(e) for e in slopehon_signal])+"\n")
signal_file.close()
norm_file.close()
slope_file.close()
# Returning normalized and slope sequences
return hon_signal, slopehon_signal
def bias_correction(self, signal, bias_table, genome_file_name, chrName, start, end):
"""
Performs bias correction.
Keyword arguments:
signal -- Input signal.
bias_table -- Bias table.
Return:
bias_corrected_signal -- Bias-corrected sequence.
"""
if(not bias_table): return signal
# Parameters
window = 50
defaultKmerValue = 1.0
# Initialization
fastaFile = Fastafile(genome_file_name)
fBiasDict = bias_table.table[0]; rBiasDict = bias_table.table[1]
k_nb = len(fBiasDict.keys()[0])
p1 = start; p2 = end
p1_w = p1 - (window/2); p2_w = p2 + (window/2)
p1_wk = p1_w - (k_nb/2); p2_wk = p2_w + (k_nb/2)
# Raw counts
nf = [0.0] * (p2_w-p1_w); nr = [0.0] * (p2_w-p1_w)
for r in self.bam.fetch(chrName, p1_w, p2_w):
if((not r.is_reverse) and (r.pos > p1_w)): nf[r.pos-p1_w] += 1.0
if((r.is_reverse) and ((r.aend-1) < p2_w)): nr[r.aend-1-p1_w] += 1.0
# Smoothed counts
Nf = []; Nr = [];
fSum = sum(nf[:window]); rSum = sum(nr[:window]);
fLast = nf[0]; rLast = nr[0]
for i in range((window/2),len(nf)-(window/2)):
Nf.append(fSum)
Nr.append(rSum)
fSum -= fLast; fSum += nf[i+(window/2)]; fLast = nf[i-(window/2)+1]
rSum -= rLast; rSum += nr[i+(window/2)]; rLast = nr[i-(window/2)+1]
# Fetching sequence
currStr = str(fastaFile.fetch(chrName, p1_wk-1, p2_wk-2)).upper()
currRevComp = AuxiliaryFunctions.revcomp(str(fastaFile.fetch(chrName,p1_wk+2, p2_wk+1)).upper())
# Iterating on sequence to create signal
af = []; ar = []
for i in range((k_nb/2),len(currStr)-(k_nb/2)+1):
fseq = currStr[i-(k_nb/2):i+(k_nb/2)]
rseq = currRevComp[len(currStr)-(k_nb/2)-i:len(currStr)+(k_nb/2)-i]
try: af.append(fBiasDict[fseq])
except Exception: af.append(defaultKmerValue)
try: ar.append(rBiasDict[rseq])
except Exception: ar.append(defaultKmerValue)
# Calculating bias and writing to wig file
fSum = sum(af[:window]); rSum = sum(ar[:window]);
fLast = af[0]; rLast = ar[0]
bias_corrected_signal = []
for i in range((window/2),len(af)-(window/2)):
nhatf = Nf[i-(window/2)]*(af[i]/fSum)
nhatr = Nr[i-(window/2)]*(ar[i]/rSum)
zf = log(nf[i]+1)-log(nhatf+1)
zr = log(nr[i]+1)-log(nhatr+1)
bias_corrected_signal.append(zf+zr)
fSum -= fLast; fSum += af[i+(window/2)]; fLast = af[i-(window/2)+1]
rSum -= rLast; rSum += ar[i+(window/2)]; rLast = ar[i-(window/2)+1]
# Termination
fastaFile.close()
return bias_corrected_signal
def hon_norm(self, sequence, mean, std):
"""
Normalizes a sequence according to hon's criterion using mean and std.
This represents a between-dataset normalization.
Keyword arguments:
sequence -- Input sequence.
mean -- Global mean.
std -- Global std.
Return:
norm_seq -- Normalized sequence.
"""
norm_seq = []
for e in sequence:
if(e == 0.0): norm_seq.append(0.0)
elif(e > 0.0): norm_seq.append(1.0/(1.0+(exp(-(e-mean)/std))))
else: norm_seq.append(-1.0/(1.0+(exp(-(-e-mean)/std))))
return norm_seq
def boyle_norm(self, sequence):
"""
Normalizes a sequence according to Boyle's criterion.
This represents a within-dataset normalization.
Keyword arguments:
sequence -- Input sequence.
Return:
norm_seq -- Normalized sequence.
"""
mean = array([e for e in sequence if e>0]).mean()
norm_seq = [(float(e)/mean) for e in sequence]
return norm_seq
def savitzky_golay_coefficients(self, window_size, order, deriv):
"""
Evaluate the Savitzky-Golay coefficients in order to evaluate the slope of the signal.
It uses a window_size (of the interpolation), order (of the polynomial), deriv (derivative needed).
Keyword arguments:
window_size -- Size of the window for function interpolation.
order -- Order of polynomial.
deriv -- Derivative.
Return:
m[::-1] -- The Savitzky-Golay coefficients.
"""
# Get statistics
#try: # TODO Errors
window_size = abs(int(window_size))
order = abs(int(order))
#except ValueError, msg:
# raise ValueError("windowSize and order have to be of type int")
#if windowSize % 2 != 1 or windowSize < 1:
# raise TypeError("windowSize size must be a positive odd number")
#if windowSize < order + 2:
# raise TypeError("windowSize is too small for the polynomials order")
order_range = range(order+1)
half_window = (window_size -1) // 2
# Precompute Coefficients
b = mat([[k**i for i in order_range] for k in range(-half_window, half_window+1)])
m = linalg.pinv(b).A[deriv]
return m[::-1]
def slope(self, sequence, sg_coefs):
"""
Evaluates the slope of sequence given the sg_coefs loaded.
Keyword arguments:
sequence -- Input sequence.
sg_coefs -- Savitzky-Golay coefficients.
Return:
slope_seq -- Slope sequence.
"""
slope_seq = convolve(sequence, sg_coefs)
slope_seq = [e for e in slope_seq[(len(sg_coefs)/2):(len(slope_seq)-(len(sg_coefs)/2))]]
return slope_seq
