def write_restraints(inp, initial_confs, start, end, tpr, top, includes, n, ndxfn, Nchains):
n = int(n) # number of points in the string, including start and end point
ndx_atoms = res_selection.read_ndx(ndxfn)
use_interpolation = False
if initial_confs is None or len(initial_confs) == 0:
use_interpolation = True
# Read the starting and ending atom configurations for later interpolation TODO
#startpts = readxvg.readxvg(start_xvg, selection)
#endpts = readxvg.readxvg(end_xvg, selection)
# Rewrite the topology to include the res itp files instead of the original per-chain itps (if any)
# There will be one topol_x.top per intermediate string point
sys.stderr.write('%s' % includes)
for k in range(n):
with open(top) as in_topf:
in_top = in_topf.read()
for mol in range(Nchains):
if len(includes) > 0:
includename = includes[mol].split('/')[-1]
in_top = re.sub(includename, 'res_%d_chain_%d.itp' % (k, mol), in_top)
with open('topol_%d.top' % k, 'w') as out_top:
# sys.stderr.write('%s'%in_top)
out_top.write(in_top)
# Generate/copy and write-out the restraint atom and force spec for each intermediate point
# This is really unnecessary here since the restraint positions are not in these files so they are the same
# for all points and chains. TODO
for k in range(n):
for mol in range(Nchains):
with open('res_%d_chain_%d.itp' % (k, mol), 'w') as restraint_itp:
if Nchains > 1:
with open(includes[mol]) as moltop_f:
moltop = moltop_f.read()
restraint_itp.write(moltop)
if len(includes) > 0:
protein = molecule(includes[mol])
# replace the chain names with the chain names
else:
with open('topol_%d.top' % k, 'w') as out_top:
protein = molecule(top)
with open(top, 'r') as in_itp_f:
in_itp = in_itp_f.read().split('; Include Position restraint file')
out_top.write(in_itp[0])
out_top.write('#include "res_%d_chain_%d.itp"\n' % (k, mol))
out_top.write(in_itp[1])
# Go through the atoms in the selection index and write one row for each one with the KFAC
# force constant placeholder
restraint_itp.write("\n[ position_restraints ]\n")
restraint_itp.write("; atom type fx fy fz\n")
for a in ndx_atoms:
if a < 5566: # GLIC HACK: only write one chain, and do it relative atom 1 since the .itp maps to the topology molecule.
restraint_itp.write("%6d 1 KFAC KFAC KFAC\n" % int(a))
def write_restraints(inp, initial_confs, start, end, start_xvg, end_xvg, tpr, top, includes, n, ndx_file, Nchains):
# Get the atoms involved with the residues to use for dihedrals (might be more than one atom in the index
# per residue, since it's probably generated by make_ndx)
ndx_atoms = res_selection.read_ndx(ndx_file)
# Map them to each affected residue so we just get the residue numbers back
selection = res_selection.res_select(start, ndx_atoms)
n = int(n) # number of points in the string, including start and end point
use_interpolation = False
if initial_confs is None or len(initial_confs) == 0:
use_interpolation = True
# Read the starting and ending dihedrals for later interpolation
startpts = readxvg.readxvg(start_xvg, selection)
endpts = readxvg.readxvg(end_xvg, selection)
else:
# Have to generate the dihedrals ourselves from the given initial structures
# Note: when we get an initial_confs[] array, we use it for all points and
# the start/end input parameters are completely ignored
# TODO: assert that len(initial_confs) == n otherwise?
ramaprocs = {}
# Run g_rama (in parallel) on each structure and output to a temporary .xvg
FNULL = open(os.devnull, 'w') # dont generate spam from g_rama
for i in range(n):
# TODO: check for and use g_rama_mpi.. like everywhere else
ramaprocs[i] = Popen(['g_rama', '-f', initial_confs[i], '-s', tpr, '-o', '0%3d.xvg' % i],
stdout=FNULL, stderr=FNULL)
# Go through the output from the rama sub-processes and read the xvg outputs
stringpts = {} # Will have 4 levels: stringpoint, residue, chain, phi/psi value
for i in range(n):
# Start array indexed by residue
xvg_i = os.path.join(inp.getOutputDir(), '0%3d.xvg' % i)
# Make sure the corresponding g_rama task has ended
ramaprocs[i].communicate()
# Read back and parse like for the start/end_xvg above
stringpts[i] = readxvg.readxvg(xvg_i, selection)
# Rewrite the topology to include the res itp files instead of the original per-chain itps (if any)
# There will be one topol_x.top per string point
sys.stderr.write('%s' % includes)
for k in range(n):
with open(top) as in_topf:
in_top = in_topf.read()
for mol in range(Nchains):
if len(includes) > 0:
includename = includes[mol].split('/')[-1]
in_top = re.sub(includename, 'res_%d_chain_%d.itp' % (k, mol), in_top)
with open('topol_%d.top' % k,'w') as out_top:
# sys.stderr.write('%s'%in_top)
out_top.write(in_top)
# Generate/copy and write-out the dihedrals for each point
for k in range(n):
for mol in range(Nchains):
# TODO: use with statement for restraint_itp as well
restraint_itp = open('res_%d_chain_%d.itp' % (k, mol), 'w')
if Nchains > 1:
with open(includes[mol]) as moltop_f:
moltop = moltop_f.read()
restraint_itp.write(moltop)
# write the initial part of the topology file
# Note: gromacs 4.6+ required
restraint_itp.write("[ dihedral_restraints ]\n")
restraint_itp.write("; ai aj ak al type phi dphi kfac\n")
if len(includes) > 0:
protein = molecule(includes[mol])
# replace the chain names with the chain names
else:
with open('topol_%d.top' % k, 'w') as out_top:
protein = molecule(top)
with open(top,'r') as in_itp_f:
in_itp = in_itp_f.read().split('; Include Position restraint file')
out_top.write(in_itp[0])
out_top.write('#include "res_%d_chain_%d.itp"\n' % (k, mol))
out_top.write(in_itp[1])
# Create a lookup-table for the protein topology that maps residue to dihedrally relevant
# backbone atom indices for N, CA and C.
dih_atoms = {}
for a in protein:
if (a.atomname == 'CA' or a.atomname == 'N' or a.atomname == 'C'):
try:
dih_atoms[a.resnr][a.atomname] = a.atomnr;
except KeyError:
dih_atoms[a.resnr] = { a.atomname: a.atomnr }
# Use the lookup-table built above and get the dihedral specification atoms needed for each
# residue in the selection. This is O(n) in residues, thanks to the dih_atoms table.
for r in selection:
# Get the atom numbers to use for the phi and psi dihedrals (4 atoms each)
# phi is C on the previous residue, and N, CA, C on this
phi = [ dih_atoms[r - 1]['C'], dih_atoms[r]['N'], dih_atoms[r]['CA'], dih_atoms[r]['C'] ]
# psi is N, CA and C on this residue and N on the next
psi = [ dih_atoms[r]['N'], dih_atoms[r]['CA'], dih_atoms[r]['C'], dih_atoms[r + 1]['N'] ]
# Write phi, psi angles and the associated k factor into a row in the restraint file
# Note: in the Gromacs 4.6+ format, the k-factor is here. Before, it was in the .mdp as
# dihre_fc.
# Also see reparametrize.py
if use_interpolation:
# k is from 0 to n-1, so map it so we get a factor from 0 to 1
phi_val = startpts[r][mol][0] + k * (endpts[r][mol][0] - startpts[r][mol][0]) / (n - 1)
psi_val = startpts[r][mol][1] + k * (endpts[r][mol][1] - startpts[r][mol][1]) / (n - 1)
else:
# Use the values extracted from the initial_confs[] structures above
phi_val = stringpts[k][r][mol][0]
psi_val = stringpts[k][r][mol][1]
# Since we need different force constants in different stages, we need to put
# a searchable placeholder in the file here and replace it later. KFAC is normally
# a %8.4f number.
restraint_itp.write("%5d%5d%5d%5d%5d %8.4f%5d KFAC\n"
%(phi[0], phi[1], phi[2], phi[3], 1, phi_val, 0))
restraint_itp.write("%5d%5d%5d%5d%5d %8.4f%5d KFAC\n"
%(psi[0], psi[1], psi[2], psi[3], 1, psi_val, 0))
restraint_itp.close()
def reparametrize(use_posres, fix_endpoints, cvs, ndx_file, Nchains,
start_conf, start_xvg, end_conf, end_xvg, last_resconfs, top,
includes):
Nswarms = len(cvs[0])
ndx_atoms = res_selection.read_ndx(ndx_file)
# For dihedrals, we map the atoms to residues for a single chain, and the readxvg etc. will read the entire file and
# select the same residues in each chain. But for the position restraints which use the atom indices directly, we have
# to first expand the index so it covers all chains.
# TODO: have to figure out or input atoms per chain in the .gro's so we can repeat the atom-selection Nchains times
# for the posres case. The ndx file is for atoms inside the chain, but the .gro will contain global numbering.
# We can detect the chain-repeat in rwgro, by looking for repeating first residue name.
# Hardcode a repeat for testing for now.
if use_posres == 0:
# Map atoms to residues for the dihedral selection
rsel = res_selection.res_select('%s' % start_conf, ndx_atoms)
#sys.stderr.write('Residue selection: %s' %rsel)
# else:
# selected_atoms = []
# for ch in range(5):
# for i in range(len(ndx_atoms)):
# selected_atoms += [ ndx_atoms[i] + ch * 5566 ]
# Calculate the average drift in CV space
# newpts is a per-swarm-point list of CV points (each a list of the CV dimension length)
newpts = []
# Note: the cvs[][] array is indexed after the number of stringpoints that actually were swarm-processed,
# so depending on the fix_endpoints option it may or may not exactly match the path[] which always include
# all points. If we only read in N-2 points here, the start/end will be added to newpts in the code further below.
for pathpt in range(len(cvs)):
swarmpts = []
for i in range(len(cvs[pathpt])):
if use_posres == 1:
zpt = rwgro.readgro_flat(cvs[pathpt][i], ndx_atoms)
#sys.stderr.write('Read pathpt %d swarm %d (%s), got %d CVs\n' % (pathpt, i, cvs[pathpt][i], len(zpt)))
else:
zpt = readxvg.readxvg_flat(cvs[pathpt][i], rsel)
swarmpts.append(zpt)
zptsum = reduce(mapadd, swarmpts)
avgdrift = scale((1 / float(Nswarms)), zptsum)
newpts.append(avgdrift)
# Read in the fixed start and end CV values, for the fix_endpoints case (otherwise the start/end will
# be allowed to drift just like the other points, and they will already then be a part of the newpts array)
if fix_endpoints == 1:
if use_posres == 1:
# TODO: the start/end_conf are full Systems so the atom numbering aliases for the ndx_atoms array :/
# Currently fixed in readgro_flat temporary, hardcoded for the GLIC Protein number.
initpt = rwgro.readgro_flat(start_conf, ndx_atoms)
targetpt = rwgro.readgro_flat(end_conf, ndx_atoms)
else:
initpt = readxvg.readxvg_flat(start_xvg, rsel)
targetpt = readxvg.readxvg_flat(end_xvg, rsel)
sys.stderr.write('Length of initpt %d, targetpt %d\n' %
(len(initpt), len(targetpt)))
# Insert the start/end in the beginning and last of newpts
newpts.insert(0, initpt)
newpts.append(targetpt)
# something with 1 indexing makes this padding necessary. TODO: check if this is needed anymore
paddingpt = [0] * len(newpts[0])
newpts.append(paddingpt)
# Do the actual reparameterization
# newpts is a 2D list, first level is one per stringpoint, second is the linear list of CVs
# rep_pts returns the maximum spread of the CV distances between points in [0] and the adjusted
# points in [1]
# Initial iteration
rep_it1 = ext_rep_pts(newpts)
adjusted = rep_it1[1] # get the points only, ignore the spread result
# Keep iterating, feeding the result of the previous result into rep_pts again
# Note that with long CV vectors (> 4000 dimensions) iterations takes a long time
# (at least 45 min for 25 iterations on a single-core 3.5 GHz) when using the python rep_pts.
# We can abort early when the maximum spread between points in the updated string goes
# below a threshold
iters = [adjusted]
i = 0
maxspread = 100.0
# Do max 150 iterations even if we don't reach our goal
while i < 150 and maxspread > 0.012:
sys.stderr.write('Rep iter %d: \n' % i)
sys.stderr.flush()
rep_it = ext_rep_pts(iters[i])
maxspread = rep_it[0]
sys.stderr.write(' maxspread was %f\n' % maxspread)
# Remember the adjusted points
iters.append(rep_it[1])
i = i + 1
sys.stderr.write('Final maximum spread %f after %d iterations.\n' %
(maxspread, i))
# Get the final iteration's result
adjusted = iters[-1]
# delete the padding point
adjusted = adjusted[:-1]
newpts = newpts[:-1]
#sys.stderr.write('Pts before repa:\n %s\n' % newpts)
#sys.stderr.write('The adjusted pts:\n %s\n' % adjusted)
# Possibility to test skipping reparametrize by uncommenting the next row.
# The stringpoints will drift along the string and probably end up in the
# endpoints or a minima along the string.
#adjusted = newpts
# calculate reparam distance
sys.stderr.write('Length of the adjusted vector: %d\n' % len(adjusted))
# TODO Nchains should depend on the specific residue (?)
# Given as function argument now.
#Nchains = len(initpt) / (2 * len(rsel))
# write the CV control data for the next iteration
# The output file expected for the posres case is rep_resconf_%d.gro for each stringpoint.
# For dihedrals its res_%d_chain_%d.itp for each stringpoint and chain.
#
for k in range(len(adjusted)):
# Not necessary to do this output for the start/end-points in the fix_endpoints case, the data is
# just bypassed in the caller script
if fix_endpoints == 1 and (k == 0 or k == (len(adjusted) - 1)):
continue
if use_posres == 1:
# Open the output resconf which will go into the next iteration as minimization target
with open('rep_resconf_%d.gro' % k, 'w') as rep_resconf:
# Open and read the previous (input) resconf, which has basically tagged along since the last
# reparametrization step (or was set initially at swarm-start)
with open(last_resconfs[k], 'r') as in_resconf_f:
in_resconf = in_resconf_f.readlines()
# TODO: maybe this chunk of code could be done by the rwgro module for us.
# Copy the first 2 rows (title and number of atoms) straight over
rep_resconf.write(in_resconf[0])
rep_resconf.write(in_resconf[1])
# Go through the atoms row-by-row and update the xyz coordinates for the atoms the reparametrize
# step moved
# Note: we are only copying over positions here. The velocities are not needed as the use for these files
# will only be as a base for the next iterations position restraint coordinates.
pathpoint = adjusted[
k] # the 1-D list of CVs (positions): x,y,z * nbr atoms in index
if len(pathpoint) != (1555 *
3): # assert on GLIC length (TODO)
sys.stderr.write('adjusted[] entry of wrong length %d\n' %
len(pathpoint))
cvpos = 0
for line in in_resconf[2:][:-1]:
resname = line[
0:
8] # python-ranges are inclusive the first index and exclusive the second...
atname = line[8:15]
atomnr = int(line[15:20])
x = float(line[20:28])
y = float(line[28:36])
z = float(line[36:44])
if atomnr in ndx_atoms:
# Update to new coords
x = pathpoint[cvpos]
y = pathpoint[cvpos + 1]
z = pathpoint[cvpos + 2]
cvpos += 3
# Write out the row, updated or not
rep_resconf.write('%s%s%5d%8.3f%8.3f%8.3f\n' %
(resname, atname, atomnr, x, y, z))
# Copy the last row which was the cell dimensions
rep_resconf.write(in_resconf[len(in_resconf) - 1])
else:
for chain in range(Nchains):
with open('res_%d_chain_%d.itp' % (k, chain),
'w') as restraint_itp:
with open(includes[k][chain], 'r') as in_itpf:
in_itp = in_itpf.read()
moltop = in_itp.split('[ dihedral_restraints ]')[0]
restraint_itp.write('%s' % moltop)
sys.stderr.write(
"Writing restraints for stringpoint %d chain %d\n" %
(k, chain))
# Note: this format is for Gromacs 4.6+
restraint_itp.write("[ dihedral_restraints ]\n")
restraint_itp.write(
"; ai aj ak al type phi dphi kfac phiB dphiB kfacB\n"
)
pathpoint = adjusted[k] # just a list of phi/psi angles
if Nchains == 1:
protein = molecule(top)
else:
protein = molecule('%s' % includes[k][chain])
# Create a lookup-table for the protein topology that maps residue to dihedrally relevant
# backbone atom indices for N, CA and C.
dih_atoms = {}
for a in protein:
if (a.atomname == 'CA' or a.atomname == 'N'
or a.atomname == 'C'):
try:
dih_atoms[a.resnr][a.atomname] = a.atomnr
except KeyError:
dih_atoms[a.resnr] = {a.atomname: a.atomnr}
# Use the lookup-table built above and get the dihedral specification atoms needed for each
# residue in the selection. This is O(n) in residues, thanks to the dih_atoms table.
pos = 0
for r in rsel:
# Get the atom numbers to use for the phi and psi dihedrals (4 atoms each)
# phi is C on the previous residue, and N, CA, C on this
phi = [
dih_atoms[r - 1]['C'], dih_atoms[r]['N'],
dih_atoms[r]['CA'], dih_atoms[r]['C']
]
# psi is N, CA and C on this residue and N on the next
psi = [
dih_atoms[r]['N'], dih_atoms[r]['CA'],
dih_atoms[r]['C'], dih_atoms[r + 1]['N']
]
# get phi and psi values from the reparametrization vector
phi_val = pathpoint[pos + chain]
psi_val = pathpoint[pos + chain + 1]
# Go to the next residue (phi,phi vals * number of chains apart)
pos += 2 * Nchains
# write phi, psi angles and k-factor
# Note: in the Gromacs 4.6+ format, the k-factor is here. Before, it was in the .mdp as
# dihre_fc.
# Since we need different force constants in different stages, we need to put
# a searchable placeholder in the file here and replace it later
restraint_itp.write(
"%5d%5d%5d%5d%5d %8.4f%5d KFAC\n" %
(phi[0], phi[1], phi[2], phi[3], 1, phi_val, 0))
restraint_itp.write(
"%5d%5d%5d%5d%5d %8.4f%5d KFAC\n" %
(psi[0], psi[1], psi[2], psi[3], 1, psi_val, 0))
def write_restraints(inp, initial_confs, start, end, start_xvg, end_xvg, tpr, top, includes, n, ndx_file, Nchains):
cmdnames = cmds.GromacsCommands()
# Get the atoms involved with the residues to use for dihedrals (might be more than one atom in the index
# per residue, since it's probably generated by make_ndx)
ndx_atoms = res_selection.read_ndx(ndx_file)
# Map them to each affected residue so we just get the residue numbers back
selection = res_selection.res_select(start, ndx_atoms)
n = int(n) # number of points in the string, including start and end point
use_interpolation = False
if initial_confs is None or len(initial_confs) == 0:
use_interpolation = True
# Read the starting and ending dihedrals for later interpolation
startpts = readxvg.readxvg(start_xvg, selection)
endpts = readxvg.readxvg(end_xvg, selection)
else:
# Have to generate the dihedrals ourselves from the given initial structures
# Note: when we get an initial_confs[] array, we use it for all points and
# the start/end input parameters are completely ignored
# TODO: assert that len(initial_confs) == n otherwise?
ramaprocs = {}
# Run g_rama (in parallel) on each structure and output to a temporary .xvg
FNULL = open(os.devnull, 'w') # dont generate spam from g_rama
for i in range(n):
# TODO: check for and use g_rama_mpi.. like everywhere else
cmd = cmdnames.rama.split() + ['-f', initial_confs[i], '-s', tpr,
'-o', '0%3d.xvg' % i]
ramaprocs[i] = Popen(cmd, stdout=FNULL, stderr=FNULL)
# Go through the output from the rama sub-processes and read the xvg outputs
stringpts = {} # Will have 4 levels: stringpoint, residue, chain, phi/psi value
for i in range(n):
# Start array indexed by residue
xvg_i = os.path.join(inp.getOutputDir(), '0%3d.xvg' % i)
# Make sure the corresponding g_rama task has ended
ramaprocs[i].communicate()
# Read back and parse like for the start/end_xvg above
stringpts[i] = readxvg.readxvg(xvg_i, selection)
# Rewrite the topology to include the res itp files instead of the original per-chain itps (if any)
# There will be one topol_x.top per string point
sys.stderr.write('%s' % includes)
for k in range(n):
with open(top) as in_topf:
in_top = in_topf.read()
for mol in range(Nchains):
if len(includes) > 0:
includename = includes[mol].split('/')[-1]
in_top = re.sub(includename, 'res_%d_chain_%d.itp' % (k, mol), in_top)
with open('topol_%d.top' % k,'w') as out_top:
# sys.stderr.write('%s'%in_top)
out_top.write(in_top)
# Generate/copy and write-out the dihedrals for each point
for k in range(n):
for mol in range(Nchains):
# TODO: use with statement for restraint_itp as well
restraint_itp = open('res_%d_chain_%d.itp' % (k, mol), 'w')
if Nchains > 1:
with open(includes[mol]) as moltop_f:
moltop = moltop_f.read()
restraint_itp.write(moltop)
# write the initial part of the topology file
# Note: gromacs 4.6+ required
restraint_itp.write("[ dihedral_restraints ]\n")
restraint_itp.write("; ai aj ak al type phi dphi kfac\n")
if len(includes) > 0:
protein = molecule(includes[mol])
# replace the chain names with the chain names
else:
with open('topol_%d.top' % k, 'w') as out_top:
protein = molecule(top)
with open(top,'r') as in_itp_f:
in_itp = in_itp_f.read().split('; Include Position restraint file')
out_top.write(in_itp[0])
out_top.write('#include "res_%d_chain_%d.itp"\n' % (k, mol))
out_top.write(in_itp[1])
# Create a lookup-table for the protein topology that maps residue to dihedrally relevant
# backbone atom indices for N, CA and C.
dih_atoms = {}
for a in protein:
if (a.atomname == 'CA' or a.atomname == 'N' or a.atomname == 'C'):
try:
dih_atoms[a.resnr][a.atomname] = a.atomnr;
except KeyError:
dih_atoms[a.resnr] = { a.atomname: a.atomnr }
# Use the lookup-table built above and get the dihedral specification atoms needed for each
# residue in the selection. This is O(n) in residues, thanks to the dih_atoms table.
for r in selection:
# Get the atom numbers to use for the phi and psi dihedrals (4 atoms each)
# phi is C on the previous residue, and N, CA, C on this
phi = [ dih_atoms[r - 1]['C'], dih_atoms[r]['N'], dih_atoms[r]['CA'], dih_atoms[r]['C'] ]
# psi is N, CA and C on this residue and N on the next
psi = [ dih_atoms[r]['N'], dih_atoms[r]['CA'], dih_atoms[r]['C'], dih_atoms[r + 1]['N'] ]
# Write phi, psi angles and the associated k factor into a row in the restraint file
# Note: in the Gromacs 4.6+ format, the k-factor is here. Before, it was in the .mdp as
# dihre_fc.
# Also see reparametrize.py
if use_interpolation:
# k is from 0 to n-1, so map it so we get a factor from 0 to 1
phi_val = startpts[r][mol][0] + k * (endpts[r][mol][0] - startpts[r][mol][0]) / (n - 1)
psi_val = startpts[r][mol][1] + k * (endpts[r][mol][1] - startpts[r][mol][1]) / (n - 1)
else:
# Use the values extracted from the initial_confs[] structures above
phi_val = stringpts[k][r][mol][0]
psi_val = stringpts[k][r][mol][1]
# Since we need different force constants in different stages, we need to put
# a searchable placeholder in the file here and replace it later. KFAC is normally
# a %8.4f number.
restraint_itp.write("%5d%5d%5d%5d%5d %8.4f%5d KFAC\n"
%(phi[0], phi[1], phi[2], phi[3], 1, phi_val, 0))
restraint_itp.write("%5d%5d%5d%5d%5d %8.4f%5d KFAC\n"
%(psi[0], psi[1], psi[2], psi[3], 1, psi_val, 0))
restraint_itp.close()
def write_restraints(inp, initial_confs, start, end, tpr, top, includes, n,
ndxfn, Nchains):
n = int(n) # number of points in the string, including start and end point
ndx_atoms = res_selection.read_ndx(ndxfn)
use_interpolation = False
if initial_confs is None or len(initial_confs) == 0:
use_interpolation = True
# Read the starting and ending atom configurations for later interpolation TODO
#startpts = readxvg.readxvg(start_xvg, selection)
#endpts = readxvg.readxvg(end_xvg, selection)
# Rewrite the topology to include the res itp files instead of the original per-chain itps (if any)
# There will be one topol_x.top per intermediate string point
sys.stderr.write('%s' % includes)
for k in range(n):
with open(top) as in_topf:
in_top = in_topf.read()
for mol in range(Nchains):
if len(includes) > 0:
includename = includes[mol].split('/')[-1]
in_top = re.sub(includename,
'res_%d_chain_%d.itp' % (k, mol), in_top)
with open('topol_%d.top' % k, 'w') as out_top:
# sys.stderr.write('%s'%in_top)
out_top.write(in_top)
# Generate/copy and write-out the restraint atom and force spec for each intermediate point
# This is really unnecessary here since the restraint positions are not in these files so they are the same
# for all points and chains. TODO
for k in range(n):
for mol in range(Nchains):
with open('res_%d_chain_%d.itp' % (k, mol), 'w') as restraint_itp:
if Nchains > 1:
with open(includes[mol]) as moltop_f:
moltop = moltop_f.read()
restraint_itp.write(moltop)
if len(includes) > 0:
protein = molecule(includes[mol])
# replace the chain names with the chain names
else:
with open('topol_%d.top' % k, 'w') as out_top:
protein = molecule(top)
with open(top, 'r') as in_itp_f:
in_itp = in_itp_f.read().split(
'; Include Position restraint file')
out_top.write(in_itp[0])
out_top.write('#include "res_%d_chain_%d.itp"\n' %
(k, mol))
out_top.write(in_itp[1])
# Go through the atoms in the selection index and write one row for each one with the KFAC
# force constant placeholder
restraint_itp.write("\n[ position_restraints ]\n")
restraint_itp.write("; atom type fx fy fz\n")
for a in ndx_atoms:
if a < 5566: # GLIC HACK: only write one chain, and do it relative atom 1 since the .itp maps to the topology molecule.
restraint_itp.write("%6d 1 KFAC KFAC KFAC\n" %
int(a))
def reparametrize(
use_posres,
fix_endpoints,
cvs,
ndx_file,
Nchains,
start_conf,
start_xvg,
end_conf,
end_xvg,
last_resconfs,
top,
includes,
):
Nswarms = len(cvs[0])
ndx_atoms = res_selection.read_ndx(ndx_file)
# For dihedrals, we map the atoms to residues for a single chain, and the readxvg etc. will read the entire file and
# select the same residues in each chain. But for the position restraints which use the atom indices directly, we have
# to first expand the index so it covers all chains.
# TODO: have to figure out or input atoms per chain in the .gro's so we can repeat the atom-selection Nchains times
# for the posres case. The ndx file is for atoms inside the chain, but the .gro will contain global numbering.
# We can detect the chain-repeat in rwgro, by looking for repeating first residue name.
# Hardcode a repeat for testing for now.
if use_posres == 0:
# Map atoms to residues for the dihedral selection
rsel = res_selection.res_select("%s" % start_conf, ndx_atoms)
# sys.stderr.write('Residue selection: %s' %rsel)
# else:
# selected_atoms = []
# for ch in range(5):
# for i in range(len(ndx_atoms)):
# selected_atoms += [ ndx_atoms[i] + ch * 5566 ]
# Calculate the average drift in CV space
# newpts is a per-swarm-point list of CV points (each a list of the CV dimension length)
newpts = []
# Note: the cvs[][] array is indexed after the number of stringpoints that actually were swarm-processed,
# so depending on the fix_endpoints option it may or may not exactly match the path[] which always include
# all points. If we only read in N-2 points here, the start/end will be added to newpts in the code further below.
for pathpt in range(len(cvs)):
swarmpts = []
for i in range(len(cvs[pathpt])):
if use_posres == 1:
zpt = rwgro.readgro_flat(cvs[pathpt][i], ndx_atoms)
# sys.stderr.write('Read pathpt %d swarm %d (%s), got %d CVs\n' % (pathpt, i, cvs[pathpt][i], len(zpt)))
else:
zpt = readxvg.readxvg_flat(cvs[pathpt][i], rsel)
swarmpts.append(zpt)
zptsum = reduce(mapadd, swarmpts)
avgdrift = scale((1 / float(Nswarms)), zptsum)
newpts.append(avgdrift)
# Read in the fixed start and end CV values, for the fix_endpoints case (otherwise the start/end will
# be allowed to drift just like the other points, and they will already then be a part of the newpts array)
if fix_endpoints == 1:
if use_posres == 1:
# TODO: the start/end_conf are full Systems so the atom numbering aliases for the ndx_atoms array :/
# Currently fixed in readgro_flat temporary, hardcoded for the GLIC Protein number.
initpt = rwgro.readgro_flat(start_conf, ndx_atoms)
targetpt = rwgro.readgro_flat(end_conf, ndx_atoms)
else:
initpt = readxvg.readxvg_flat(start_xvg, rsel)
targetpt = readxvg.readxvg_flat(end_xvg, rsel)
sys.stderr.write("Length of initpt %d, targetpt %d\n" % (len(initpt), len(targetpt)))
# Insert the start/end in the beginning and last of newpts
newpts.insert(0, initpt)
newpts.append(targetpt)
# something with 1 indexing makes this padding necessary. TODO: check if this is needed anymore
paddingpt = [0] * len(newpts[0])
newpts.append(paddingpt)
# Do the actual reparameterization
# newpts is a 2D list, first level is one per stringpoint, second is the linear list of CVs
# rep_pts returns the maximum spread of the CV distances between points in [0] and the adjusted
# points in [1]
# Initial iteration
rep_it1 = ext_rep_pts(newpts)
adjusted = rep_it1[1] # get the points only, ignore the spread result
# Keep iterating, feeding the result of the previous result into rep_pts again
# Note that with long CV vectors (> 4000 dimensions) iterations takes a long time
# (at least 45 min for 25 iterations on a single-core 3.5 GHz) when using the python rep_pts.
# We can abort early when the maximum spread between points in the updated string goes
# below a threshold
iters = [adjusted]
i = 0
maxspread = 100.0
# Do max 150 iterations even if we don't reach our goal
while i < 150 and maxspread > 0.012:
sys.stderr.write("Rep iter %d: \n" % i)
sys.stderr.flush()
rep_it = ext_rep_pts(iters[i])
maxspread = rep_it[0]
sys.stderr.write(" maxspread was %f\n" % maxspread)
# Remember the adjusted points
iters.append(rep_it[1])
i = i + 1
sys.stderr.write("Final maximum spread %f after %d iterations.\n" % (maxspread, i))
# Get the final iteration's result
adjusted = iters[-1]
# delete the padding point
adjusted = adjusted[:-1]
newpts = newpts[:-1]
# sys.stderr.write('Pts before repa:\n %s\n' % newpts)
# sys.stderr.write('The adjusted pts:\n %s\n' % adjusted)
# Possibility to test skipping reparametrize by uncommenting the next row.
# The stringpoints will drift along the string and probably end up in the
# endpoints or a minima along the string.
# adjusted = newpts
# calculate reparam distance
sys.stderr.write("Length of the adjusted vector: %d\n" % len(adjusted))
# TODO Nchains should depend on the specific residue (?)
# Given as function argument now.
# Nchains = len(initpt) / (2 * len(rsel))
# write the CV control data for the next iteration
# The output file expected for the posres case is rep_resconf_%d.gro for each stringpoint.
# For dihedrals its res_%d_chain_%d.itp for each stringpoint and chain.
#
for k in range(len(adjusted)):
# Not necessary to do this output for the start/end-points in the fix_endpoints case, the data is
# just bypassed in the caller script
if fix_endpoints == 1 and (k == 0 or k == (len(adjusted) - 1)):
continue
if use_posres == 1:
# Open the output resconf which will go into the next iteration as minimization target
with open("rep_resconf_%d.gro" % k, "w") as rep_resconf:
# Open and read the previous (input) resconf, which has basically tagged along since the last
# reparametrization step (or was set initially at swarm-start)
with open(last_resconfs[k], "r") as in_resconf_f:
in_resconf = in_resconf_f.readlines()
# TODO: maybe this chunk of code could be done by the rwgro module for us.
# Copy the first 2 rows (title and number of atoms) straight over
rep_resconf.write(in_resconf[0])
rep_resconf.write(in_resconf[1])
# Go through the atoms row-by-row and update the xyz coordinates for the atoms the reparametrize
# step moved
# Note: we are only copying over positions here. The velocities are not needed as the use for these files
# will only be as a base for the next iterations position restraint coordinates.
pathpoint = adjusted[k] # the 1-D list of CVs (positions): x,y,z * nbr atoms in index
if len(pathpoint) != (1555 * 3): # assert on GLIC length (TODO)
sys.stderr.write("adjusted[] entry of wrong length %d\n" % len(pathpoint))
cvpos = 0
for line in in_resconf[2:][:-1]:
resname = line[0:8] # python-ranges are inclusive the first index and exclusive the second...
atname = line[8:15]
atomnr = int(line[15:20])
x = float(line[20:28])
y = float(line[28:36])
z = float(line[36:44])
if atomnr in ndx_atoms:
# Update to new coords
x = pathpoint[cvpos]
y = pathpoint[cvpos + 1]
z = pathpoint[cvpos + 2]
cvpos += 3
# Write out the row, updated or not
rep_resconf.write("%s%s%5d%8.3f%8.3f%8.3f\n" % (resname, atname, atomnr, x, y, z))
# Copy the last row which was the cell dimensions
rep_resconf.write(in_resconf[len(in_resconf) - 1])
else:
for chain in range(Nchains):
with open("res_%d_chain_%d.itp" % (k, chain), "w") as restraint_itp:
with open(includes[k][chain], "r") as in_itpf:
in_itp = in_itpf.read()
moltop = in_itp.split("[ dihedral_restraints ]")[0]
restraint_itp.write("%s" % moltop)
sys.stderr.write("Writing restraints for stringpoint %d chain %d\n" % (k, chain))
# Note: this format is for Gromacs 4.6+
restraint_itp.write("[ dihedral_restraints ]\n")
restraint_itp.write("; ai aj ak al type phi dphi kfac phiB dphiB kfacB\n")
pathpoint = adjusted[k] # just a list of phi/psi angles
if Nchains == 1:
protein = molecule(top)
else:
protein = molecule("%s" % includes[k][chain])
# Create a lookup-table for the protein topology that maps residue to dihedrally relevant
# backbone atom indices for N, CA and C.
dih_atoms = {}
for a in protein:
if a.atomname == "CA" or a.atomname == "N" or a.atomname == "C":
try:
dih_atoms[a.resnr][a.atomname] = a.atomnr
except KeyError:
dih_atoms[a.resnr] = {a.atomname: a.atomnr}
# Use the lookup-table built above and get the dihedral specification atoms needed for each
# residue in the selection. This is O(n) in residues, thanks to the dih_atoms table.
pos = 0
for r in rsel:
# Get the atom numbers to use for the phi and psi dihedrals (4 atoms each)
# phi is C on the previous residue, and N, CA, C on this
phi = [dih_atoms[r - 1]["C"], dih_atoms[r]["N"], dih_atoms[r]["CA"], dih_atoms[r]["C"]]
# psi is N, CA and C on this residue and N on the next
psi = [dih_atoms[r]["N"], dih_atoms[r]["CA"], dih_atoms[r]["C"], dih_atoms[r + 1]["N"]]
# get phi and psi values from the reparametrization vector
phi_val = pathpoint[pos + chain]
psi_val = pathpoint[pos + chain + 1]
# Go to the next residue (phi,phi vals * number of chains apart)
pos += 2 * Nchains
# write phi, psi angles and k-factor
# Note: in the Gromacs 4.6+ format, the k-factor is here. Before, it was in the .mdp as
# dihre_fc.
# Since we need different force constants in different stages, we need to put
# a searchable placeholder in the file here and replace it later
restraint_itp.write(
"%5d%5d%5d%5d%5d %8.4f%5d KFAC\n" % (phi[0], phi[1], phi[2], phi[3], 1, phi_val, 0)
)
restraint_itp.write(
"%5d%5d%5d%5d%5d %8.4f%5d KFAC\n" % (psi[0], psi[1], psi[2], psi[3], 1, psi_val, 0)
)
