def ATANPIItv( self, a):
aa=str(a)
if (aa=="nan;nan"):
return "nan;nan"
elif aa.count(";")==0:
b1=down(lambda: float(a))
b2=up(lambda: float(a))
res1=atanpi(interval([b1,b2]))
res2=str(interval.hull([res1]))
aa=res2.split("[")[1]
bb=aa.split(",")[0]
cc=aa.split(" ")[1]
dd=cc.split("]")[0]
return bb +";"+dd
else:
a1=down(lambda: float(a.split(";")[0]))
a2=up(lambda: float(a.split(";")[1]))
c=interval([a1,a2])
res=atanpi(c)
res1=str(interval.hull([res]))
aa=res1.split("[")[1]
bb=aa.split(",")[0]
cc=aa.split(" ")[1]
dd=cc.split("]")[0]
return bb +";"+dd
def test_cosh(self):
self.assertEqual(imath.cosh(0), interval[1])
assert imath.cosh(2) in (imath.exp(2) + imath.exp(-2))/2
assert imath.cosh(interval[1, 2]) == interval.hull((imath.cosh(1), imath.cosh(2)))
assert imath.cosh(interval[-2, -1]) == imath.cosh(interval[1, 2])
assert imath.cosh(interval[-2, 1]) == interval.hull((interval[1], imath.cosh(2)))
assert imath.cosh(interval[-1, 2]) == interval.hull((interval[1], imath.cosh(2)))
def test_cosh(self):
self.assertEqual(imath.cosh(0), interval[1])
assert imath.cosh(2) in (imath.exp(2) + imath.exp(-2)) / 2
assert imath.cosh(interval[1, 2]) == interval.hull(
(imath.cosh(1), imath.cosh(2)))
assert imath.cosh(interval[-2, -1]) == imath.cosh(interval[1, 2])
assert imath.cosh(interval[-2, 1]) == interval.hull(
(interval[1], imath.cosh(2)))
assert imath.cosh(interval[-1, 2]) == interval.hull(
(interval[1], imath.cosh(2)))
def __init__(self, x, y): ## or use interval.hull self.a = interval.hull((x.a, y.a)) self.b = interval.hull((x.b, y.b)) self.real_max = max(max(self.a)) self.real_min = min(min(self.a)) self.im_max = max(max(self.b)) self.im_min = min(min(self.b))
def phi(x):
def g1(x):
return x[0] * imath.sin(x[0]) + 0.1 * (x[0]**2) + 1.0
def g2(x):
return imath.cos(x[1]) + 0.1 * (x[1]**2)
res = [g1(x), g2(x)]
phi = interval[max(interval.hull([interval(u) for u in x])[0][0] for x in res),\
max(interval.hull([interval(u) for u in x])[0][1] for x in res)]
return phi[0][0], phi[0][1]
def __step(self, box):
dom = box[self.__v_name]
f = self.fun.eval(box)
#print('f: '+str(f))
if is_empty(f) or not is_superset(f, 0):
return interval()
df = self.fun.d_eval(self.__v_id, box)
#print('df: '+str(df))
if is_empty(df):
return interval()
c = self.sample_fun(box[self.__v_name])
# TODO
box[self.__v_name] = interval[c]
fc = self.fun.eval(box)
#print('fc: '+str(fc))
box[self.__v_name] = dom
l, r = ext_div(fc, df)
#print('l: '+str(l)+', r: '+str(r))
l = c - l
r = c - r
l &= dom
r &= dom
if is_empty(l):
return r
else:
return interval.hull([l, r])
def contract(self, box):
op, l, r = self.__constr
l = l[0]
r = r[0]
# forward propagation
self.__fwd_eval(l, box)
self.__fwd_eval(r, box)
print('fwd:')
print(self.__fwd)
print()
# backward propagation
fwd = self.__fwd
bwd = self.__bwd
if op == '=':
v = fwd[l] & fwd[r]
bwd[l] = v
self.__bwd_propag(l, box)
bwd[r] = v
self.__bwd_propag(r, box)
elif op == '>' or op == '>=':
v = interval.hull([fwd[r], interval[inf]])
bwd[l] = fwd[l] & v
self.__bwd_propag(l, box)
v = interval.hull([interval[-inf], fwd[l]])
bwd[r] = fwd[r] & v
self.__bwd_propag(r, box)
elif op == '<' or op == '<=':
v = interval.hull([interval[-inf], fwd[r]])
bwd[l] = fwd[l] & v
self.__bwd_propag(l, box)
v = interval.hull([fwd[l], interval[inf]])
bwd[r] = fwd[r] & v
self.__bwd_propag(r, box)
print('bwd:')
print(self.__bwd)
print()
def bound_root(self, x):
diff = self.diffPoly()
bdiff = diff.bound_naive(x)
if bdiff[0].inf >= 0 or bdiff[0].sup <= 0:
return interval[interval(self(x[0].inf)), interval(self(x[0].sup))]
else:
diff2 = diff.diffPoly()
roots = x.newton(diff, diff2)
bound = interval[interval(self(x[0].inf)),
interval(self(x[0].sup))]
for i in roots:
bound = interval.hull((bound, self(interval(i))))
return bound
def element_inv(n, x0, D):
coef = []
for i in range(n + 1):
coef.append(Taylor.coeff('inv', i, x0, 1))
P = Poly(coef, n)
coef1 = Taylor.coeff('inv', n + 1, D, 1)
if coef1[0].inf >= 0 or coef1[0].sup <= 0:
a = interval(D[0].inf)
b = interval(D[0].sup)
I1 = interval(1) / a - P.bound(a - x0)
I2 = interval(1) / b - P.bound(b - x0)
I0 = interval(1) / interval(x0) - P.bound(interval(0))
I = (interval.hull((I1, I2, I0))) & (coef1 * ((D - x0)**(n + 1)))
else:
I = coef1 * ((D - x0)**(n + 1))
return Taylor(P, I, x0, D)
def element(f, n, x0, D):
coef = []
for i in range(n + 1):
coef.append(Taylor.coeff(f, i, x0, 1))
P = Poly(coef, n)
coef1 = Taylor.coeff(f, n + 1, D, 1)
f = eval('imath.' + f)
if coef1[0].inf >= 0 or coef1[0].sup <= 0:
a = interval(D[0].inf)
b = interval(D[0].sup)
I1 = f(a) - P.bound_best(a - x0, ['n', 'H'])
I2 = f(b) - P.bound_best(b - x0, ['n', 'H'])
I0 = f(interval(x0)) - P.bound(interval(0))
I = (interval.hull((I1, I2, I0))) & (coef1 * ((D - x0)**(n + 1)))
else:
I = coef1 * ((D - x0)**(n + 1))
return Taylor(P, I, x0, D)
def __shrink_upper(self, box):
self.__newton.sample_fun = sample_sup
vn = self.__v_name
old = box[vn]
while True:
self.__newton.contract(box)
#print(box)
if self.__is_consistent_u(box) or is_empty(box[vn]):
# restore the lb
box[vn] = interval.hull([interval[old[0].inf, box[vn]]])
return
else:
dom = box[vn]
box[vn] = interval[dom.midpoint, dom[0].inf]
def UNIONItv( self, *args ):
from interval import interval
somme=interval()
c=str(args)
c=c.replace(",,",",")
c=c.replace(")","")
c=c.replace("(","")
c=c.replace(",,",",")
c=c.replace("'","")
c=c.replace(" ","")
c=c.replace("''","")
c=c.replace(",,",",")
mm=c.split(",")
i=0
while (i<len(mm)): #< len(mm)
cour=str(mm[i])
if cour != "None" and cour !="":
cour=cour.replace("'","")
cour=cour.replace(" ","")
cour=cour.replace(",","")
if cour=="nan;nan":
somme=somme
elif cour.count(";")==0:
b1=down(lambda: float(cour))
b2=up(lambda: float(cour))
from interval import interval, inf, imath
somme=somme|interval([b1,b2])
else:
a1=down(lambda:float(cour.split(";")[0]))
a2=up(lambda:float(cour.split(";")[1]))
#from interval import interval
somme=somme|interval([a1,a2])
somme=interval.hull([somme])
i+=1
res=str(somme)
if res.count(",")==0:
res1=res.split("[")[1]
res2=res1.split("]")[0]
return str(res2 )
else:
aa=res.split("[")[1]
bb=aa.split(",")[0]
cc=aa.split(" ")[1]
dd=cc.split("]")[0]
return bb +";"+dd
def DIVItv( self, a, b ):
if a=="nan;nan" or b=="nan;nan":
return "nan;nan"
else:
if a.count(";")==0 and b.count(";")==0:
a1=down(lambda: float(a))
b1=down(lambda: float(b))
a2=up(lambda: float(a))
b2=up(lambda: float(b))
elif a.count(";")==0:
a1=down(lambda: float(a))
a2=up(lambda: float(a))
b1=down(lambda: float(b.split(";")[0]))
b2=up(lambda: float(b.split(";")[1]))
elif b.count(";")==0:
a1=down(lambda: float(a.split(";")[0]))
b1=down(lambda: float(b))
a2=up(lambda: float(a.split(";")[1]))
b2=up(lambda: float(b))
else:
a1=down(lambda: float(a.split(";")[0]))
b1=down(lambda: float(b.split(";")[0]))
a2=up(lambda: float(a.split(";")[1]))
b2=up(lambda: float(b.split(";")[1]))
from interval import interval
inv=interval([b1,b2])
invcomplet=inv.inverse()
res=interval([a1,a2])*invcomplet
res1=str(interval.hull([res]))
if res1.count(" ")==0:
aa=res1.split("[")[1]
dd=aa.split("]")[0]
return dd +";"+dd
else:
aa=res1.split("[")[1]
bb=aa.split(",")[0]
cc=aa.split(" ")[1]
dd=cc.split("]")[0]
return bb +";"+dd
def ABSItv( self, a):
aa=str(a)
if (aa=="nan;nan"):
return "nan;nan"
elif aa.count(";")==0:
a1=down(lambda: float(aa))
a2=up(lambda: float(aa))
else:
a1=down(lambda: float(a.split(";")[0]))
a2=up(lambda: float(a.split(";")[1]))
c=interval([a1,a2])
res=c.__abs__()
res1=str(interval.hull([res]))
if res1.count(" ")==0:
aa=res1.split("[")[1]
dd=aa.split("]")[0]
return dd +";"+dd
else:
aa=res1.split("[")[1]
bb=aa.split(",")[0]
cc=aa.split(" ")[1]
dd=cc.split("]")[0]
return bb +";"+dd
def test_hull(self):
self.assertEqual(
interval([1, 9]),
interval.hull((interval([1, 3], [4, 6]), interval([2, 5], 9))))
def test_hull(self):
self.assertEqual(interval([1, 9]), interval.hull((interval([1, 3], [4, 6]), interval([2, 5], 9))))
def __bwd_propag(self, n_id, box):
n = self.dag[n_id]
fwd = self.__fwd
bwd = self.__bwd
rec = self.__bwd_propag
if n[0] == '+':
bwd[n[1]] = bwd[n_id] - fwd[n[2]]
rec(n[1], box)
bwd[n[2]] = bwd[n_id] - fwd[n[1]]
rec(n[2], box)
elif n[0] == '-':
bwd[n[1]] = bwd[n_id] + fwd[n[2]]
rec(n[1], box)
bwd[n[2]] = fwd[n[1]] - bwd[n_id]
rec(n[2], box)
elif n[0] == '*':
bwd[n[1]] = bwd[n_id] / fwd[n[2]]
rec(n[1], box)
bwd[n[2]] = bwd[n_id] / fwd[n[1]]
rec(n[2], box)
elif n[0] == '/':
bwd[n[1]] = bwd[n_id] * fwd[n[2]]
rec(n[1], box)
bwd[n[2]] = fwd[n[1]] / bwd[n_id]
rec(n[2], box)
elif n[0] == '^':
i = self.dag[n[2]][1]
if i % 2 == 0:
p = root(bwd[n_id], i)
pp = p & fwd[n[1]]
np = (-p) & fwd[n[1]]
if is_empty(pp) or is_empty(np):
bwd[n[1]] = interval()
else:
bwd[n[1]] = interval.hull([pp, np])
else:
bwd[n[1]] = root(bwd[n_id], i)
rec(n[1], box)
elif n[0] == 'sqrt':
if is_empty(bwd[n_id]) or bwd[n_id][0].sup < 0:
bwd[n[1]] = interval()
elif bwd[n_id][0].inf < 0:
i = interval([0, bwd[n_id][0].sup])
bwd[n[1]] &= i * i
else:
bwd[n[1]] &= bwd[n_id] * bwd[n_id]
assert (not is_empty(bwd[n[1]]))
# TODO
#elif n[0] == 'sin':
elif n[0] == 'C':
bwd[n_id] &= n[1]
elif n[0] == 'V':
box[n[1]] &= bwd[n_id]
else:
print('unsupported node: ' + str(n))
assert (False)
def test_hull(self):
assert interval([1, 9]) == interval.hull((interval([1, 3], [4, 6]), interval([2, 5], 9)))
def mutant_locations(self, loc, count=1, max=False):
'''
this function takes from the dict of all potential _mutation_locations a
set of mutations that fall within a loc tuple. It returns an iterator that
spits out potential mutants at these locations; it is randomized by position
first, then by mutation.
count is the number of mutations to return, setting to one returns all
possible sequences off by one, setting to two returns all sequences with
two mutations made, etc, etc.
'''
#first, make sure my self._mutant_locations dict is instantiatied
if not hasattr(self, '_mutant_locations'):
self._mutant_locations = mutate.mutant_locations(self)
#deal with interval() versus tuple inputs
if isinstance(loc, interval):
if len(loc) == 0:
return iter([])
loc_ivl = loc
loc = interval.hull([loc_ivl]).to_tuple()
else:
loc_ivl = interval(loc)
#create an iterator that returns all keys for _mutant_locations that are
#in this location range
mut_ivls = (interval(ml) for ml in self._mutant_locations.keys())
#now loc iter will output a non-random set of mutation locations which are
#keys to the _mutation_locations dict
loc_iter = ifilter(itemgetter(1), ((ivl, ivl.overlaps(loc_ivl)) for \
ivl in mut_ivls))
#change interval obj into loc tuple
loc_tup = lambda loc: loc[0].to_tuple()
#get the mutation set (the values) for a loc tuple
loc_muts = lambda loc: self._mutant_locations[loc_tup(loc)]
#expand the mutation set into individual mutations for a loc tuple
loc_mset = lambda loc: ((loc_tup(loc), i) for i in loc_muts(loc))
#put them all together for a randomized list of generators, one generator
#for each loc tuple
pos_mut_sets = map(lambda loc: (loc_mset(loc)), loc_iter)
emit_sets = combinations(util.irandomize(
chain.from_iterable(pos_mut_sets), seed=random_seed), count)
emit_sets = imap(frozenset, emit_sets)
is_unique_pos = \
lambda mset: (
len([m[0] for m in mset]) == len(set([m[0] for m in mset]))
and set(mset) not in self.mut_sets)
mut_iter = util.irandomize(
ifilter(is_unique_pos,
util.irandomize(emit_sets,
seed=random_seed)),
seed=random_seed)
# if this feature overlaps exons
#expand the motif to codons, so that we can check that mutants are
# synonymous
if interval(self.exon_list[0].extract_pos()).overlaps(loc_ivl):
codon_loc = \
(interval(mutate.expand_motif_to_codons(self, loc)) \
& interval(self.exon_list[0].extract_pos())).to_tuple()
#check all mutations for synonymousness
seq_str = str(self.seq)[slice(*codon_loc)]
is_synon = lambda seq_str, codon_loc: lambda mut_tups: \
mutate.check_translation(\
string.upper(mutate.tups_to_str(seq_str, codon_loc, mut_tups)),
seq_str)
is_synon = is_synon(seq_str, codon_loc)
return util.irandomize(ifilter(lambda mut: is_synon(mut), mut_iter),
seed=random_seed)
else:
return mut_iter
def im_radius(self): """ Returns the radius of this complex interval's imaginary part """ hull = interval.hull((self.b, self.b)) return max(max(hull)) - min(min(hull))
def test_hull(self):
assert interval([1, 9]) == interval.hull((interval([1, 3], [4, 6]), interval([2, 5], 9)))
def gis(bedfiles, names=None, prefix="similarity", sim_thresh=0.5):
"""
Calculate genomic similarity of BED files
Parameters
----------
beds : str
Path to BED files to compare
names : List<str> or NoneType
Name for each of the input BED files
prefix : str
Output file prefix
sim_thresh : float
Minimum similarity threshold to consider recording a locus
"""
# file handles for each BED file
n = len(bedfiles)
# store sample indexes as `names` is `names` is not defined
if names is None:
names = ["Sample_" + str(i) for i in range(n)]
# store similarity matrix
sim_mat = np.identity(n)
# columns of information to store
column_names = ["chr", "start", "end"] + ["similarity"] + names
# initialize `bed` objects
beds = [Bed(b) for b in bedfiles]
# get first intervals from each file
intvls = [b.next() for b in beds]
# initialize the matrix
for i, j in combinations(range(n), 2):
sim_mat[i, j] = similarity(intvls[i], intvls[j])
sim_mat = symmetrize(sim_mat)
# store minimum similarity for each interval, and which sample it comes from
minsim = [{
"idx": j,
"s": sim_mat[i, j]
} for i, j in enumerate(np.argmin(sim_mat, axis=1))]
# records to keep for printing
records = []
# iterate over intervals from the sorted BED files
while True:
minsim_set = min([ms["s"] for ms in minsim])
chrom = intvls[0].chr
# skip if not all intervals are on the same chromosome
# find interval that spans entire set of intervals
if np.all([intvl.chr == chrom for intvl in intvls]):
# record this set if the similarity of the set passes the threshold
if minsim_set >= sim_thresh:
hull = interval.hull([intvl.interval for intvl in intvls])
set_locus = GenomicInterval(chrom, hull[0].inf, hull[0].sup)
records.append(
dict((colname, v) for colname, v in zip(
column_names,
[
set_locus.chr,
set_locus.inf,
set_locus.sup,
minsim_set,
*[b.counter for b in beds],
],
)))
else:
# update to latest chromosome and skip remaining intervals
# get latest chromosome (via `natsorted` to account for chromosome names)
newchrom = natsorted([intvl.chr for intvl in intvls])[-1]
for i in range(n):
if intvls[i].chr == newchrom:
skip_idx = i
break
# keep track of which samples to update
samples_to_update = list(range(skip_idx)) + list(
range(skip_idx + 1, n))
for i in samples_to_update:
while True:
# iterate through intervals in BED files until all samples are on the new chromosome
intvls[i] = beds[i].next()
if intvls[i].chr == newchrom:
break
# find sample with the smallest upper bound
update_idx = np.argmin([intvl.sup for intvl in intvls])
# pop this interval
intvls[update_idx] = beds[update_idx].next()
# check that we're not at the end of the file
if intvls[update_idx] is None:
break
# recalculate column of sim_mat (calc once, ensure sim_mat is symmetric)
for i in range(update_idx):
sim_mat[i, update_idx] = similarity(intvls[i], intvls[update_idx])
sim_mat[update_idx, i] = sim_mat[i, update_idx]
for i in range(update_idx + 1, n):
sim_mat[i, update_idx] = similarity(intvls[i], intvls[update_idx])
sim_mat[update_idx, i] = sim_mat[i, update_idx]
# update minsim for (update_idx)-th sample
minsim[update_idx]["idx"] = np.argmin(sim_mat[:, update_idx])
minsim[update_idx]["s"] = sim_mat[update_idx,
minsim[update_idx]["idx"]]
# update minsim for any sample where minsim[j]["idx"] == update_idx
for i in [i for i, ms in enumerate(minsim) if ms["idx"] == update_idx]:
minsim[i]["idx"] = np.argmin(sim_mat[:, i])
minsim[i]["s"] = sim_mat[i, minsim[i]["idx"]]
# save records as a DataFrame
df = pd.DataFrame(records, columns=column_names)
# save to output
df.to_csv(prefix + ".tsv", index=False, sep="\t")
return df
def _hull(w, z): """ Returns the hull of two complex intervals """ return ComplexInterval(interval.hull((w.a, z.a)), interval.hull((w.b, z.b)))
def mutate_all_positions(self, loc):
'''
mutate every codon and/or nucleotide within feature bounds
'''
#first, make sure my self._mutant_locations dict is instantiated
if not hasattr(self, '_mutant_locations'):
self._mutant_locations = mutate.mutant_locations(self)
#deal with interval() versus tuple inputs
if isinstance(loc, interval):
loc_ivl = loc
loc = interval.hull([loc_ivl]).to_tuple()
else:
loc_ivl = interval(loc)
ivl_len = loc_ivl.sum_len()
(e_coords, i_coords) = mutate.get_motif_boundaries(loc, self)
#mutant choices will be a list of random.choice lambda functions
#that randomly chooses a different codon for every position or a different
#nucleotide for every intronic base
mutant_choices = set()
for codon_loc in e_coords:
#go through every codon in codon_loc
for c_loc in range(codon_loc[0], codon_loc[1], 3):
codon = str(self.seq[c_loc:(c_loc + 3)]).upper()
#get other codons
bckt = mutate.codon_back_table()
fwdt = mutate.codon_fwd_table()
other_codons = bckt[fwdt[codon]]
other_codons = other_codons.difference((codon,))
if len(other_codons) == 0:
continue
#convert these codons into mut tuples (cmut_tuples)
# (one codon might be two or even three tuples)
cmut_tuples = ()
for other_cod in other_codons:
cod_tup = ()
for diff in util.str_diff(other_cod, codon):
diff_loc = (c_loc + diff, c_loc + diff + 1)
cod_tup += ((diff_loc, other_cod[diff]),)
cmut_tuples += (cod_tup,)
#finally store a lambda function that randomly chooses a
#different codon for this position, using a unique-state
#random generator
rgen = random.Random()
rgen.seed(random_seed ^ hash(cmut_tuples) ^ hash(loc))
codon_choice = lambda cmt, rgen: lambda: rgen.choice(cmt)
mutant_choices.add(codon_choice(cmut_tuples, rgen))
for intron_loc in i_coords:
intron_ivl = interval(intron_loc)
mut_ivls = (interval(ml) for ml in self._mutant_locations.keys())
loc_list = filter(itemgetter(1), \
[(ivl, ivl.overlaps(intron_ivl)) for ivl in mut_ivls])
#change interval obj into loc tuple
loc_tup = lambda loc: loc[0].to_tuple()
#get the mutation set (the values) for a loc tuple
loc_muts = lambda loc: self._mutant_locations[loc_tup(loc)]
#expand the mutation set into individual mutations for a loc tuple
loc_mset = lambda loc, rgen: \
lambda: rgen.choice([((loc_tup(loc), i),) for i in loc_muts(loc)])
#generate independently seeded random number gens for each pos
rgens = [random.Random() for i in loc_list]
[rg.seed((random_seed, loc)) for rg, loc in zip(rgens, loc_list)]
pos_rnd_muts = map(lambda loc, rgen: loc_mset(loc, rgen), loc_list, rgens)
mutant_choices.update(pos_rnd_muts)
#now that we have a mutant choices list with one function for every
# codon/nt, we need to create a generator that calls each function in the
# list once only
while True:
yielded = set()
next_mut = frozenset(
chain.from_iterable(map(lambda f: f(), mutant_choices)))
seen_count = 0
if next_mut not in yielded:
yielded.add(next_mut)
yield next_mut
elif next_mut in yielded and seen_count < 20:
seen_count += 1
elif next_mut in yielded and seen_count >= 20:
raise StopIteration
def re_radius(self): """ Returns the radius of this complex interval's real part """ hull = interval.hull((self.a, self.a)) return max(max(hull)) - min(min(hull))