def rmsd(struct1, struct2):
coords1 = struct1.getCoords()
coords2 = struct2.getCoords()
assert len(coords1) == len(coords2)
ssd = 0
ssd = sum([
la.calcDistance(coord1, coord2)
for coord1, coord2 in zip(coords1, coords2)
])
msd = ssd / len(coords1)
return msd**(1. / 2)
def error(dists, coords):
assert len(dists) == len(coords)
n = len(dists)
sse = 0
count = 0
for i in range(n):
for j in range(i):
embedded_dist = la.calcDistance(coords[i], coords[j])
sse += (embedded_dist - dists[i, j])**2
count += 1
mse = sse / count
rmse = mse**(1. / 2)
return rmse
mat1 = dt.matFromBed(
"hic_data/GM12878_combined_{}_{}kb.bed".format(chrom, res_kb), struct1)
comps1 = ca.get_compartments(mat1, struct1)
mat2 = dt.matFromBed("hic_data/K562_{}_{}kb.bed".format(chrom, res_kb),
struct2)
comps2 = ca.get_compartments(mat2, struct2)
r, p = st.pearsonr(comps1, comps2)
if r < 0:
comps1 = -comps1
comp_diffs = np.abs(comps1 - comps2)
dists = np.array([
la.calcDistance(coord1, coord2)
for coord1, coord2 in zip(struct1.getCoords(), struct2.getCoords())
])
dist_peaks = sg.find_peaks_cwt(dists, np.arange(1, 10))
plt.subplot2grid((10, 10), (0, 0), 9, 10, frameon=False)
gen_coords = struct1.getGenCoords()
plt.plot(gen_coords,
comp_diffs / max(comp_diffs),
lw=2,
color=(0.75, 0, 0),
label="Compartment score change",
zorder=1)
plt.plot(gen_coords,
dists / max(dists),
lw=2,
#load structures
structure1 = dt.structure_from_file("hic_data/{}_{}_{}kb_structure.tsv".format(cell_type1, chrom, res_kb))
structure2 = dt.structure_from_file("hic_data/{}_{}_{}kb_structure.tsv".format(cell_type2, chrom, res_kb))
#rescale
structure1.rescale()
structure2.rescale()
#make structures compatible
dt.make_compatible((structure1, structure2))
#calculate changes
coords1 = np.array(structure1.getCoords())
coords2 = np.array(structure2.getCoords())
dists = [la.calcDistance(coord1, coord2) for coord1, coord2 in zip(coords1, coords2)]
#compartments
chrom1 = dt.chromFromBed(path1)
chrom2 = dt.chromFromBed(path2)
chrom1.res = 100000 #reduce res to reduce RAM usage in compartment calculation
chrom2.res = 100000
chrom1.minPos = int(np.floor(float(chrom1.minPos)/chrom1.res)) * chrom1.res #round
chrom1.maxPos = int(np.ceil(float(chrom1.maxPos)/chrom1.res)) * chrom1.res
chrom2.minPos = int(np.floor(float(chrom2.minPos)/chrom2.res)) * chrom2.res #round
chrom2.maxPos = int(np.ceil(float(chrom2.maxPos)/chrom2.res)) * chrom2.res
low_struct1 = dt.structureFromBed(path1, chrom1)
low_struct2 = dt.structureFromBed(path2, chrom2)
dt.make_compatible((low_struct1, low_struct2))
contacts1 = dt.matFromBed(path1, low_struct1)
ps = np.arange(0, 0.1, 0.01)
for p in ps:
all_changes = []
for i in range(n):
multimds.full_mds(path1, path2, penalty=p)
structure1 = dt.structure_from_file("{}_structure.tsv".format(os.path.basename(prefix1)))
structure2 = dt.structure_from_file("{}_structure.tsv".format(os.path.basename(prefix2)))
if p == 0:
r, t = la.getTransformation(structure1, structure2)
structure1.transform(r,t)
all_changes.append(np.array([la.calcDistance(coord1, coord2) for coord1, coord2 in zip(structure1.getCoords(), structure2.getCoords())]))
r_sq = []
for i in range(n):
for j in range(i):
r, p = st.pearsonr(all_changes[i], all_changes[j])
r_sq.append(r**2)
all_r_sq.append(r_sq)
ys = all_r_sq
#start with a frameless plot (extra room on the left)
plt.subplot2grid((10,10), (0,0), 9, 10, frameon=False)
#label axes
import matplotlib
from matplotlib import pyplot as plt
from multimds import multimds
from multimds import linear_algebra as la
struct1, struct2 = multimds.full_mds("sim1_chr21_100kb.bed", "sim2_chr21_100kb.bed")
gen_coords = struct1.getGenCoords()
dists = [la.calcDistance(coord1, coord2) for coord1, coord2 in zip(struct1.getCoords(), struct2.getCoords())]
plt.subplot2grid((10,10), (0,0), 9, 10, frameon=False)
plt.plot(gen_coords, dists, lw=2)
#define offsets
xmin = min(gen_coords)
xmax = max(gen_coords)
x_range = xmax - xmin
x_start = xmin - x_range/25.
x_end = xmax + x_range/25.
ymin = 0
ymax = max(dists)
y_range = ymax - ymin
y_start = ymin - y_range/25.
y_end = ymax + y_range/25.
#define axes with offsets
plt.axis([x_start, x_end, y_start, y_end], frameon=False)
#plot axes (black with line width of 4)
plt.axvline(x=x_start, color="k", lw=4)