def plot_GC(chr,tbx_gc,cp_vect,starts,ends):
F=open("GC.txt",'w')
cp_vect = cp_vect.astype(np.float64)
var = get_windowed_variance(cp_vect,50)
F.write("var\tgc\n")
for i in xrange(50,starts.shape[0],101):
s = starts[i-50]
e = ends[i+50]
gc = np.mean(np.array([float(l[3]) for l in tbx_gc.fetch(chr,s,e,parser=pysam.asTuple())]))
if var[i] != 0:
print >>F,"%f\t%f"%( var[i],gc )
exit(1)
def init_all_dists(self):
self.get_indiv_window_dist()
self.csum_all = np.cumsum(self.all_cps)
self.mu_probDensByWidth = {}
self.ll_probDensByWidth = {}
self.half_width = 250
self.variance_vect = get_windowed_variance(self.all_cps, self.half_width)
self.var_left=np.roll(self.variance_vect,self.half_width+1)
self.var_right=np.roll(self.variance_vect,-self.half_width-1)
"""
var_left and var_right represent at position k,
the variance of the 2*half_width+1 windows to the right
and the left
"""
self.get_variance_dist()
self.null_var = np.var(self.all_cps)
self.null_mu = np.mean(self.all_cps)
self.initialized = True
def init_all_dists(self):
self.get_indiv_window_dist()
self.csum_all = np.cumsum(self.all_cps)
self.mu_probDensByWidth = {}
self.ll_probDensByWidth = {}
self.half_width = 250
self.variance_vect = get_windowed_variance(self.all_cps, self.half_width)
self.var_left=np.roll(self.variance_vect,self.half_width+1)
self.var_right=np.roll(self.variance_vect,-self.half_width-1)
"""
var_left and var_right represent at position k,
the variance of the 2*half_width+1 windows to the right
and the left
"""
self.get_variance_dist()
self.null_var = np.var(self.all_cps)
self.null_mu = np.mean(self.all_cps)
self.initialized = True
def __init__(self,chr,cp_data, starts, ends, cutoff_scale, **kwargs):
max_merge=kwargs.get("max_merge",0.5)
use_means=kwargs.get("use_means",False)
n_scales=kwargs.get("n_scales",51)
#n_scales=kwargs.get("n_scales",30)
scale_width=kwargs.get('scale_width',1)
n_bin_smoothings=kwargs.get('-n_bin_smoothings',0)
smoothing_kernel=kwargs.get('smoothing_kernel',np.array([1,2,1]))
self.chr = chr
self.cutoff_scale = cutoff_scale
self.scales = list(np.arange(1,n_scales,scale_width))
self.starts = starts
self.ends = ends
self.n_wnds = self.starts.shape[0]
self.cp_data = cp_data
self.der1=np.zeros((len(self.scales),self.n_wnds),dtype=np.float32)
self.der2=np.zeros((len(self.scales),self.n_wnds),dtype=np.int8)
self.vars = get_windowed_variance(cp_data.astype(np.float64),500)
self.l_vars = np.roll(self.vars,501)
self.r_vars = np.roll(self.vars,-501)
print >>stderr, "scales range from %f-%f"%(self.scales[0],self.scales[-1])
for i in xrange(n_bin_smoothings):
print >>stderr,"doing binomial smooth #%d"%i
cp_data=ndi.convolve1d(cp_data,smoothing_kernel)/np.sum(smoothing_kernel)
transitions_by_scale = {}
print >>stderr, "finding contours..."
for i_scale,scale in enumerate(self.scales):
stderr.write("%.2f "%(scale))
stderr.flush()
g1=ndi.gaussian_filter1d(cp_data,scale,order=1)
g2=ndi.gaussian_filter1d(cp_data,scale,order=2)
edges,pos_edges,neg_edges = self.get_n_edges(g1,g2)
self.der1[i_scale,:]=g1
self.der2[i_scale,:]=pos_edges-neg_edges
transitions_by_scale[scale]=(edges,pos_edges,neg_edges)
stderr.write("done\n")
self.contour_intersects,x_intercept_to_scale=get_contours(self.der2)
######NOW we have all the per-scale contours
#print contour_intersects
edges_passing_cutoff =[]
curr_all_edges=[]
curr_all_edges_scales=[]
#take all the edges discovered at some scale
for scale,edges in self.contour_intersects.iteritems():
curr_all_edges.extend(edges)
curr_all_edges_scales.extend([scale for i in xrange(len(edges))])
if scale >=cutoff_scale:
edges_passing_cutoff.extend(edges)
edges_passing_cutoff=sorted(set(edges_passing_cutoff))
all_edges_scales=sorted(zip(curr_all_edges,curr_all_edges_scales))
stderr.write("hierarchically merging segments\n")
t = time.time()
segments_s, segments_e, cps = c_hierarch_merge_edges(cp_data,
edges_passing_cutoff,
max_merge,
use_means,
self.n_wnds,
self.starts,
self.ends)
#segments_s, segments_e, cps = hierarch_merge_edges(cp_data,
# edges_passing_cutoff,
# max_merge,use_means)
self.segment_edges=(segments_s,segments_e,cps)
print >>stderr, "hierarchical clustering completed in %fs"%(time.time()-t)
def __init__(self, chr, cp_data, starts, ends, cutoff_scale, **kwargs):
max_merge = kwargs.get("max_merge", 0.5)
use_means = kwargs.get("use_means", False)
n_scales = kwargs.get("n_scales", 51)
#n_scales=kwargs.get("n_scales",30)
scale_width = kwargs.get('scale_width', 1)
n_bin_smoothings = kwargs.get('-n_bin_smoothings', 0)
smoothing_kernel = kwargs.get('smoothing_kernel', np.array([1, 2, 1]))
self.chr = chr
self.cutoff_scale = cutoff_scale
self.scales = list(np.arange(1, n_scales, scale_width))
self.starts = starts
self.ends = ends
self.n_wnds = self.starts.shape[0]
self.cp_data = cp_data
self.der1 = np.zeros((len(self.scales), self.n_wnds), dtype=np.float32)
self.der2 = np.zeros((len(self.scales), self.n_wnds), dtype=np.int8)
self.vars = get_windowed_variance(cp_data.astype(np.float64), 500)
self.l_vars = np.roll(self.vars, 501)
self.r_vars = np.roll(self.vars, -501)
print("scales range from %f-%f" % (self.scales[0], self.scales[-1]),
file=stderr)
for i in range(n_bin_smoothings):
print("doing binomial smooth #%d" % i, file=stderr)
cp_data = ndi.convolve1d(
cp_data, smoothing_kernel) / np.sum(smoothing_kernel)
transitions_by_scale = {}
print("finding contours...", file=stderr)
for i_scale, scale in enumerate(self.scales):
stderr.write("%.2f " % (scale))
stderr.flush()
g1 = ndi.gaussian_filter1d(cp_data, scale, order=1)
g2 = ndi.gaussian_filter1d(cp_data, scale, order=2)
edges, pos_edges, neg_edges = self.get_n_edges(g1, g2)
self.der1[i_scale, :] = g1
self.der2[i_scale, :] = pos_edges - neg_edges
transitions_by_scale[scale] = (edges, pos_edges, neg_edges)
stderr.write("done\n")
self.contour_intersects, x_intercept_to_scale = get_contours(self.der2)
######NOW we have all the per-scale contours
#print contour_intersects
edges_passing_cutoff = []
curr_all_edges = []
curr_all_edges_scales = []
#take all the edges discovered at some scale
for scale, edges in self.contour_intersects.items():
curr_all_edges.extend(edges)
curr_all_edges_scales.extend([scale for i in range(len(edges))])
if scale >= cutoff_scale:
edges_passing_cutoff.extend(edges)
edges_passing_cutoff = sorted(set(edges_passing_cutoff))
all_edges_scales = sorted(zip(curr_all_edges, curr_all_edges_scales))
stderr.write("hierarchically merging segments\n")
t = time.time()
segments_s, segments_e, cps = c_hierarch_merge_edges(
cp_data, edges_passing_cutoff, max_merge, use_means, self.n_wnds,
self.starts, self.ends)
#segments_s, segments_e, cps = hierarch_merge_edges(cp_data,
# edges_passing_cutoff,
# max_merge,use_means)
self.segment_edges = (segments_s, segments_e, cps)
print("hierarchical clustering completed in %fs" % (time.time() - t),
file=stderr)
