def overlap_plot(self, receptor, ligand, T, name='overlap'):
import _Volume
vt1 = receptor.sum(dim=0)
vt2 = ligand.sum(dim=0)
_Volume.Volume2Xplor(vt1.cpu(), "receptor_%s.xplor" % name,
self.resolution)
_Volume.Volume2Xplor(vt2.cpu(), "ligand_%s.xplor" % name,
self.resolution)
dx = int(T[0])
dy = int(T[1])
dz = int(T[2])
L = self.box_size
def mabs(dx, dy, dz):
import numpy as np
return np.abs(dx), np.abs(dy), np.abs(dz)
#all positive
if dx > 0 and dy > 0 and dz > 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[dx:L, dy:L, dz:L] += vt2[0:L - dx, 0:L - dy, 0:L - dz]
#one negative
elif dx < 0 and dy > 0 and dz > 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[0:L - dx, dy:L, dz:L] += vt2[dx:L, 0:L - dy, 0:L - dz]
elif dx > 0 and dy < 0 and dz > 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[dx:L, 0:L - dy, dz:L] += vt2[0:L - dx, dy:L, 0:L - dz]
elif dx > 0 and dy > 0 and dz < 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[dx:L, dy:L, 0:L - dz] += vt2[0:L - dx, 0:L - dy, dz:L]
#one positive
elif dx < 0 and dy < 0 and dz > 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[0:L - dx, 0:L - dy, dz:L] += vt2[dx:L, dy:L, 0:L - dz]
elif dx > 0 and dy < 0 and dz < 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[dx:L, 0:L - dy, 0:L - dz] += vt2[0:L - dx, dy:L, dz:L]
elif dx < 0 and dy > 0 and dz < 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[0:L - dx, dy:L, 0:L - dz] += vt2[dx:L, 0:L - dy, dz:L]
#all negative
elif dx < 0 and dy < 0 and dz < 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[0:L - dx, 0:L - dy, 0:L - dz] += vt2[dx:L, dy:L, dz:L]
_Volume.Volume2Xplor(vt1.cpu(), "%s.xplor" % name, self.resolution)
def runTest(self):
volume_gpu = self.tc2v(self.coords, self.num_atoms_of_type, self.offsets)
volume = volume_gpu.sum(dim=1).to(device='cpu', dtype=torch.float)
if not os.path.exists('TestFig'):
os.mkdir('TestFig')
_Volume.Volume2Xplor(volume[0,:,:,:], "TestFig/total_b0_vtest_%d_%.1f.xplor"%(self.box_size, self.resolution), self.resolution)
def plot_double_volume(self, volume, name='volume'):
import _Volume
L = volume.size(1)
halfL = int(L / 2)
doubleL = L * 2
output_volume = torch.zeros(doubleL,
doubleL,
doubleL,
device='cpu',
dtype=torch.float)
output_volume[L - halfL:L + halfL, L - halfL:L + halfL,
L - halfL:L + halfL] = volume[0:L, 0:L, 0:L]
_Volume.Volume2Xplor(output_volume, "%s.xplor" % name, self.resolution)
def plot_circ_volume(self, circ_volume, name='volume'):
import _Volume
L = circ_volume.size(1)
L2 = int(L / 2)
output_volume = torch.zeros(L, L, L, device='cpu', dtype=torch.float)
# quadrants 1 <--> 8
output_volume[:L2, :L2, :L2] = circ_volume[L2:L, L2:L, L2:L]
output_volume[L2:L, L2:L, L2:L] = circ_volume[:L2, :L2, :L2]
# quadrants 2 <--> 9
output_volume[L2:L, :L2, :L2] = circ_volume[:L2, L2:L, L2:L]
output_volume[:L2, L2:L, L2:L] = circ_volume[L2:L, :L2, :L2]
# quadrants 3 <--> 5
output_volume[L2:L, L2:L, :L2] = circ_volume[:L2, :L2, L2:L]
output_volume[:L2, :L2, L2:L] = circ_volume[L2:L, L2:L, :L2]
# quadrants 4 <--> 6
output_volume[:L2, L2:L, :L2] = circ_volume[L2:L, :L2, L2:L]
output_volume[L2:L, :L2, L2:L] = circ_volume[:L2, L2:L, :L2]
_Volume.Volume2Xplor(output_volume, "%s.xplor" % name, self.resolution)
inp1_border = torch.zeros(1, 1, v_size, v_size, v_size, dtype=torch.float, device='cuda') inp2 = torch.zeros(1, 1, v_size, v_size, v_size, dtype=torch.float, device='cuda') out2 = torch.zeros(1, 1, v_size, v_size, v_size, dtype=torch.float, device='cuda') fill_V1(np.array(r0), np.array(r1), 8, 5, inp1) get_boundary(inp1, inp1_border) fill_V2(np.array(l0), 5, inp2) # v = conv_numpy(inp1, inp2) # xyz = (0, 0, 0) # image1 = np.concatenate((v[xyz[0],:,:], v[:,xyz[1],:], v[:,:,xyz[2]]), axis = 1) # plt.imshow(image1) # plt.colorbar() # plt.show() _Volume.Volume2Xplor(inp1.squeeze().cpu(), "receptor.xplor", 1.0) _Volume.Volume2Xplor(inp1_border.squeeze().cpu(), "receptor_border.xplor", 1.0) _Volume.Volume2Xplor(inp2.squeeze().cpu(), "ligand.xplor", 1.0) border_overlap = vc(inp1_border, inp2) bulk_overlap = vc(inp1, inp2) print(bulk_overlap.size(), torch.max(bulk_overlap), torch.min(bulk_overlap)) out = border_overlap - 0.5*bulk_overlap _Volume.Volume2Xplor(out.squeeze().cpu(), "out.xplor", 1.0) score, r = get_argmax(out[0,0,:,:,:]) print(r) r = list(r) if r[0]>=v_size: r[0] = -(2*v_size - r[0]) if r[1]>=v_size:
resolution = 1.0
sequence = ['GGAGRRRGGWG']
angles = torch.zeros(len(sequence), 7, len(sequence[0]), dtype=torch.double)
angles[0,0,:] = -1.047
angles[0,1,:] = -0.698
angles[0,2:,:] = 110.4*np.pi/180.0
a2c = Angles2Coords()
c2cc = Coords2CenteredCoords(rotate=False, translate=False, box_size=box_size, resolution=resolution)
c2tc = Coords2TypedCoords()
tc2v = TypedCoords2Volume(box_size=box_size, resolution=resolution)
protein, res_names, atom_names, num_atoms = a2c(angles, sequence)
protein = c2cc(protein, num_atoms)
coords, num_atoms_of_type, offsets = c2tc(protein, res_names, atom_names, num_atoms)
volume = tc2v(coords.cuda(), num_atoms_of_type.cuda(), offsets.cuda())
# print(volume.size())
# for i in range(0, res_names.size(1)):
# print(res_names.data[0,i,:].numpy().astype(dtype=np.uint8).tostring().split('\0')[0], atom_names.data[0,i,:].numpy().astype(dtype=np.uint8).tostring().split('\0')[0])
for i in range(0,11):
print(num_atoms_of_type.data[0,i], offsets.data[0,i])
volume = volume.sum(dim=1).squeeze().cpu()
print(volume.size())
if not os.path.exists('TestFig'):
os.mkdir('TestFig')
_Volume.Volume2Xplor(volume.squeeze(), "TestFig/total_vtest_%d_%.1f.xplor"%(box_size, resolution), resolution)
def test_gradient():
vc = VolumeConvolution()
v_size = 30
inp1 = torch.zeros(1,
1,
v_size,
v_size,
v_size,
dtype=torch.float,
device='cuda',
requires_grad=True)
inp1tmp_p = torch.zeros(1,
1,
v_size,
v_size,
v_size,
dtype=torch.float,
device='cuda')
inp1tmp_m = torch.zeros(1,
1,
v_size,
v_size,
v_size,
dtype=torch.float,
device='cuda')
inp2 = torch.zeros(1,
1,
v_size,
v_size,
v_size,
dtype=torch.float,
device='cuda',
requires_grad=True)
target = torch.zeros(1,
1,
2 * v_size,
2 * v_size,
2 * v_size,
dtype=torch.float,
device='cuda')
fill_V2(np.array([12, 30, 50]), 5, target)
r0 = [20, 15, 15]
r1 = [23, 15, 7]
l0 = [7, 24, 7]
fill_V1(np.array(r0), np.array(r1), 8, 5, inp1)
fill_V2(np.array(l0), 5, inp2)
_Volume.Volume2Xplor(inp1.squeeze().cpu(), "receptor.xplor", 1.0)
_Volume.Volume2Xplor(inp2.squeeze().cpu(), "ligand.xplor", 1.0)
out = vc(inp1, inp2)
E = torch.sum((out - target) * (out - target))
# E = torch.sum(torch.abs(out-target))
# E = out[:,:,8,40,0] + out[:,:,10,30,25]
E.backward()
back_grad_x0 = torch.zeros(inp2.grad.size(),
dtype=torch.float,
device='cpu').copy_(inp2.grad)
print("Grad max min", torch.max(back_grad_x0), torch.min(back_grad_x0))
_Volume.Volume2Xplor(back_grad_x0[0, 0, :, :, :], "grad_conv.xplor", 1.00)
num_grads = []
an_grads = []
for j in range(0, v_size):
x = j
y = 7
z = 7
dx = 0.1
inp1tmp_p.copy_(inp2)
inp1tmp_m.copy_(inp2)
inp1tmp_p[0, 0, x, y, z] += dx
inp1tmp_m[0, 0, x, y, z] -= dx
# out1_p = vc(inp1tmp_p, inp2)
# out1_m = vc(inp1tmp_m, inp2)
out1_p = vc(inp1, inp1tmp_p)
out1_m = vc(inp1, inp1tmp_m)
E1_p = torch.sum((out1_p - target) * (out1_p - target))
E1_m = torch.sum((out1_m - target) * (out1_m - target))
# E1_p = torch.sum(torch.abs(out1_p-target))
# E1_m = torch.sum(torch.abs(out1_m-target))
# E1 = (out1[:,:,8,40,0] + out1[:,:,10,30,25])
dy_dx = (E1_p.item() - E1_m.item()) / (2 * dx)
num_grads.append(dy_dx)
an_grads.append(back_grad_x0[0, 0, x, y, z].item())
fig = plt.figure()
plt.plot(num_grads, 'r.-', label='num grad')
plt.plot(an_grads, 'bo', label='an grad')
plt.legend()
plt.savefig('TestFig/test_backward.png')
labels = labels.squeeze()
print(receptor[0])
print(ligand[0])
print(labels[0])
print(labels)
vt1 = volume1[0, :, :, :, :].squeeze()
vt2 = volume2[0, :, :, :, :].squeeze()
vt1 = vt1.sum(dim=0)
vt2 = vt2.sum(dim=0)
dx = int(dT[0, 0])
dy = int(dT[0, 1])
dz = int(dT[0, 2])
_Volume.Volume2Xplor(vt1.cpu(), "receptor.xplor", 1.5)
_Volume.Volume2Xplor(vt2.cpu(), "ligand.xplor", 1.5)
print(dx, dy, dz)
L = 80
def mabs(dx, dy, dz):
import numpy as np
return np.abs(dx), np.abs(dy), np.abs(dz)
#all positive
if dx > 0 and dy > 0 and dz > 0:
dx, dy, dz = mabs(dx, dy, dz)
vt1[dx:L, dy:L, dz:L] += vt2[0:L - dx, 0:L - dy, 0:L - dz]
#one negative
angles.data[:, 0, :] = -1.047
angles.data[:, 1, :] = -0.698
angles.data[:, 2:, :] = 110.4 * np.pi / 180.0
#Converting angles to coordinates
coords, res_names, atom_names, num_atoms = a2c(angles, sequences)
#Translating the structure to fit inside the volume
translation = torch.tensor([[60, 60, 60]],
dtype=torch.double,
device='cpu')
coords = translate(coords, translation, num_atoms)
#Converting to typed coords manually
#We need dimension 1 equal to 11, because right now 11 atom types are implemented and
#this layer expects 11 offsets and 11 num atoms of type per each structure
coords = coords.to(dtype=torch.double, device='cuda')
num_atoms_of_type = torch.zeros(1, 11, dtype=torch.int, device='cuda')
num_atoms_of_type.data[0, 0] = num_atoms.data[0]
offsets = torch.zeros(1, 11, dtype=torch.int, device='cuda')
offsets.data[:, 1:] = num_atoms.data[0]
#Projecting to volume
tc2v = Volume.TypedCoords2Volume(box_size=120)
volume = tc2v(coords, num_atoms_of_type, offsets)
#Saving the volume to a file, we need to sum all the atom types, because
#Volume2Xplor currently accepts only 3d volumes
volume = volume.sum(dim=1).to(device='cpu').squeeze()
_Volume.Volume2Xplor(volume, "volume.xplor")
#Generating protein and initial volume
protein, res_names, atom_names, num_atoms = a2c(angles, sequence)
protein = c2cc(protein, num_atoms)
coords, num_atoms_of_type, offsets = c2tc(protein, res_names, atom_names, num_atoms)
volume = tc2v(coords.cuda(), num_atoms_of_type.cuda(), offsets.cuda())
#Rotating protein
center = torch.zeros(len(sequence), 3, dtype=torch.double, device='cpu').fill_((box_size - 0.5)*resolution/2.0)
centered_coords = translate(coords, -center, num_atoms)
rotated_centered_coords = rotate(centered_coords, R, num_atoms)
rotated_coords = translate(rotated_centered_coords, center, num_atoms)
#Reprojecting protein
volume_protRot = tc2v(rotated_coords.cuda(), num_atoms_of_type.cuda(), offsets.cuda())
#Rotating volume
volume_rot = volume_rotate(volume, R.to(dtype=torch.float, device='cuda'))
err = torch.sum(torch.abs(volume_protRot - volume_rot))/torch.sum(volume_rot)
print(err)
volume = volume.sum(dim=1).squeeze().cpu()
volume_protRot = volume_protRot.sum(dim=1).squeeze().cpu()
volume_rot = volume_rot.sum(dim=1).squeeze().cpu()
if not os.path.exists('TestFig'):
os.mkdir('TestFig')
_Volume.Volume2Xplor(volume_protRot.squeeze(), "TestFig/total_vPRot_%d_%.1f.xplor"%(box_size, resolution), resolution)
_Volume.Volume2Xplor(volume_rot.squeeze(), "TestFig/total_vRot_%d_%.1f.xplor"%(box_size, resolution), resolution)
_Volume.Volume2Xplor(volume.squeeze(), "TestFig/total_v_%d_%.1f.xplor"%(box_size, resolution), resolution)
angles = torch.zeros(len(sequence),
7,
len(sequence[0]),
dtype=torch.double)
angles[0, 0, :] = -1.047
angles[0, 1, :] = -0.698
angles[0, 2:, :] = 110.4 * np.pi / 180.0
a2c = Angles2Coords()
c2cc = Coords2CenteredCoords()
c2tc = Coords2TypedCoords()
tc2v = TypedCoords2Volume()
protein, res_names, atom_names, num_atoms = a2c(angles, sequence)
protein = c2cc(protein, num_atoms)
coords, num_atoms_of_type, offsets = c2tc(protein, res_names, atom_names,
num_atoms)
volume = tc2v(coords.cuda(), num_atoms_of_type.cuda(), offsets.cuda())
# print(volume.size())
# for i in range(0, res_names.size(1)):
# print(res_names.data[0,i,:].numpy().astype(dtype=np.uint8).tostring().split('\0')[0], atom_names.data[0,i,:].numpy().astype(dtype=np.uint8).tostring().split('\0')[0])
for i in range(0, 11):
print(num_atoms_of_type.data[0, i], offsets.data[0, i])
volume = volume.sum(dim=1).squeeze().cpu()
print(volume.size())
if not os.path.exists('TestFig'):
os.mkdir('TestFig')
_Volume.Volume2Xplor(volume.squeeze(), "TestFig/total_vtest.xplor")
