sidechain modelling performance benchmark

sidechain/.~lock.eval_promod_frm_no_subrotamer_optimization.csv#

+,schdaude,schdaudoputer,29.06.2020 09:45,file:///home/schdaude/.config/libreoffice/4;
\ No newline at end of file

sidechain/README.md

+Scripts and data to reproduce the sidechain modelling accuracy benchmark.
+
+The whole benchmark is a three step process:
+
+1. Fetch Benchmark Set
+2. Modelling with ProMod3 and SCWRL4
+3. Evaluation
+
+Fetch Benchmark Set
+-------------------
+
+Structures for benchmarking are fetched from the pdb using the remote loading
+capabilities from OpenStructure:
+
+`ost get_scwrl_testset_structures.py`
+
+loads, cleans and saves structures to the scwrl_testset directory.
+
+
+Modelling with ProMod3 and SCWRL4
+---------------------------------
+
+`pm reconstruct_sidechains_promod.py` reconstructs sidechains using different
+different settings: 
+
+- Flexible Rotamer Model (FRM) with subrotamer optimization with default 
+  (backbone dependent) rotamer library
+- FRM without subrotamer optimization with default rotamer library
+- Rigid Rotamer Model (RRM) with default rotamer library
+- FRM with subrotamer optimization and backbone indepenedent rotamer library
+- RRM with backbone indepenedent rotamer library
+
+The generated models appear in the subdirectories of models/promod
+
+`python reconstruct_sidechains_scwrl.py` reconstructs sidechains with
+the flexible rotamer model as well as the rigid rotamer model. The script
+requires the Scwrl4 executable in your path and the generated models appear
+in the subdirectories of models/scwrl
+
+Evaluation
+----------
+
+`pm do_plot.py` recreates Fig 2 from ProMod3 manuscript.
+
+`pm do_supplemental_tables.py` creates detailled csv tables with performances
+of the alternative modelling runs from the last step.
+

sidechain/analyze.py

+
+import os
+import numpy as np
+import traceback
+from promod3 import sidechain
+from ost import io, seq, geom
+
+def _GetAmbigRMSD(r_one, r_two, fix_atoms, ambig_atom_pairs):
+
+    rmsd_one = 0.0
+    rmsd_two = 0.0
+
+    for aname in fix_atoms:
+        a = r_one.FindAtom(aname)
+        b = r_two.FindAtom(aname)
+        if a.IsValid() and b.IsValid():
+            dist = geom.Distance(a.GetPos(),b.GetPos())
+            rmsd_one += dist*dist
+            rmsd_two += dist*dist
+        else:
+            return float('NaN')
+
+    # process rmsd_one: everything matches nicely
+    for anames in ambig_atom_pairs:
+        a = r_one.FindAtom(anames[0])
+        b = r_two.FindAtom(anames[0])
+        if not (a.IsValid() and b.IsValid()):
+            return float('NaN')        
+        dist = geom.Distance(a.GetPos(),b.GetPos())
+        rmsd_one += dist*dist
+        a = r_one.FindAtom(anames[1])
+        b = r_two.FindAtom(anames[1])
+        if not (a.IsValid() and b.IsValid()):
+            return float('NaN')        
+        dist = geom.Distance(a.GetPos(),b.GetPos())
+        rmsd_one += dist*dist
+
+    # process rmsd_two: the ambig atom pairs are flipped
+    for anames in ambig_atom_pairs:
+        a = r_one.FindAtom(anames[0])
+        b = r_two.FindAtom(anames[1])
+        if not (a.IsValid() and b.IsValid()):
+            return float('NaN')        
+        dist = geom.Distance(a.GetPos(),b.GetPos())
+        rmsd_two += dist*dist
+        a = r_one.FindAtom(anames[1])
+        b = r_two.FindAtom(anames[0])
+        if not (a.IsValid() and b.IsValid()):
+            return float('NaN')        
+        dist = geom.Distance(a.GetPos(),b.GetPos())
+        rmsd_two += dist*dist
+
+    n_atoms = len(fix_atoms) + 2 * len(ambig_atom_pairs)
+    return min(np.sqrt(rmsd_one / n_atoms), np.sqrt(rmsd_two / n_atoms))
+
+
+
+def GetRMSD(r_one, r_two):
+
+    rname = r_one.GetName()
+    if rname != r_two.GetName():
+        return float('NaN')
+    # handle the ambiguous cases
+    if rname == 'TYR':
+        return _GetAmbigRMSD(r_one, r_two, ['CG', 'CZ', 'OH'], 
+                             [('CD1', 'CD2'), ('CE1', 'CE2')])
+    if rname == 'VAL':
+        return _GetAmbigRMSD(r_one, r_two, [], [('CG1', 'CG2')])
+    if rname == 'PHE':
+        return _GetAmbigRMSD(r_one, r_two, ['CG', 'CZ'],
+                             [('CD1', 'CD2'), ('CE1', 'CE2')])
+    if rname == 'ASP':
+        return _GetAmbigRMSD(r_one, r_two, ['CG'], [('OD1', 'OD2')])
+    if rname == 'GLU':
+        return _GetAmbigRMSD(r_one, r_two, ['CG', 'CD'], [('OE1', 'OE2')])
+    # the general case
+    rm = 0.0
+    counter = 0
+    for a in r_one.atoms:
+        b = r_two.FindAtom(a.GetName())
+        if a.IsValid() and b.IsValid():
+            dist = geom.Distance(a.GetPos(),b.GetPos())
+            rm += dist*dist
+            counter+=1
+        else:
+            return float('NaN')
+    return np.sqrt(rm/counter)
+
+
+class AAEval:
+  def __init__(self):
+    self.num_valid_chi1 = 0.0
+    self.num_correct_chi1 = 0.0
+    self.num_valid_chi2 = 0.0
+    self.num_correct_chi2 = 0.0
+    self.num_valid_chi2_given_chi1 = 0.0
+    self.num_correct_chi2_given_chi1 = 0.0
+    self.rmsd_values = list()
+    self.num_residues = 0.0
+
+
+def GetAAEvaluations(model_dir, target_dir = 'scwrl_testset', 
+                     angle_thresh =  20.0/180.0*np.pi):
+
+    AA = ['ARG','ASN','ASP','CYS','GLN','GLU','HIS','ILE','LEU','LYS','MET','PHE',
+          'PRO','SER','THR','TRP','TYR','VAL']
+    eval_data = {aa : AAEval() for aa in AA}
+    testset_ids = set(f[:4] for f in os.listdir(target_dir))
+    for target_idx, t in enumerate(testset_ids):
+        try:
+            target_path = os.path.join(target_dir, t + '-molcked.pdb')
+            model_path = os.path.join(model_dir, t + '.pdb')
+            target_structure = io.LoadPDB(target_path).Select('peptide=true')
+            model_structure = io.LoadPDB(model_path).Select('peptide=true')
+            # If several chains, we only consider unique sequences of chains
+            # => first occurence matters (e.g. in case of homo oligomers)
+            present_chains = list()
+            present_sequences = list()
+            for ch in target_structure.chains:
+                ch_s = seq.CreateSequence('A', ''.join([r.one_letter_code for r in ch.residues]))
+                already_there = False
+                for s in present_sequences:
+                    aln = seq.alg.GlobalAlign(ch_s, s, seq.alg.BLOSUM62)[0]
+                    s_id = seq.alg.SequenceIdentity(aln)
+                    if s_id > 0.9:
+                        # High sequence identity can be caused through a supper crappy 
+                        # alignment where we mostly have gaps. Lets estimate the fraction 
+                        # of aligned residues for the shorter sequence
+                        s0 = str(aln.GetSequence(0))
+                        s1 = str(aln.GetSequence(1))
+                        n_aligned = sum([int(a!='-' and b!='-') for a,b in zip(s0, s1)])
+                        frac = float(n_aligned) / min(sum([int(c!='-') for c in s0]),
+                                                      sum([int(c!='-') for c in s1]))
+                        # set arbitrary threshold, most residues are expected to be aligned
+                        if frac > 0.8:
+                            already_there = True
+                if not already_there:
+                    present_chains.append(ch.GetName())
+                    present_sequences.append(ch_s)
+            relevant_query = 'cname=' + ','.join(present_chains)
+            relevant_target = target_structure.Select(relevant_query)
+            residues_to_consider = relevant_target.residues
+            for r_t in residues_to_consider:
+                rot_id = sidechain.TLCToRotID(r_t.GetName())
+                if rot_id == sidechain.XXX or rot_id == sidechain.GLY or rot_id == sidechain.ALA:
+                    continue
+                r_m = model_structure.FindResidue(r_t.GetChain().GetName(),r_t.GetNumber())
+                if not r_m.IsValid():
+                    continue
+                try:
+                    rot_t = sidechain.RotamerLibEntry.FromResidue(r_t)
+                    rot_m = sidechain.RotamerLibEntry.FromResidue(r_m)
+                except:
+                    continue # one of the residues is incomplete
+                eval_data[r_t.GetName()].num_residues += 1
+                t_chi1 = rot_t.chi1
+                m_chi1 = rot_m.chi1
+                t_chi2 = rot_t.chi2
+                m_chi2 = rot_m.chi2
+                chi1_ok = rot_t.SimilarDihedral(rot_m, 0, angle_thresh, rot_id)
+                chi2_ok = rot_t.SimilarDihedral(rot_m, 1, angle_thresh, rot_id)
+                if t_chi1 == t_chi1 and m_chi1 == m_chi1:
+                    eval_data[r_t.GetName()].num_valid_chi1 += 1
+                    if chi1_ok:
+                        eval_data[r_t.GetName()].num_correct_chi1 += 1
+                if t_chi2 == t_chi2 and m_chi2 == m_chi2:
+                    eval_data[r_t.GetName()].num_valid_chi2 += 1
+                    if chi2_ok:
+                        eval_data[r_t.GetName()].num_correct_chi2 += 1
+                if chi1_ok:
+                    if t_chi2 == t_chi2 and m_chi2 == m_chi2:
+                        eval_data[r_t.GetName()].num_valid_chi2_given_chi1 += 1
+                        if chi2_ok:
+                            eval_data[r_t.GetName()].num_correct_chi2_given_chi1 += 1
+                rm = GetRMSD(r_t, r_m)
+                if rm == rm:
+                    eval_data[r_t.GetName()].rmsd_values.append(rm)
+        except:
+            print('failed to evaluate target', t)
+            traceback.print_exc()
+
+    return eval_data
+
+
+def PrintPerformance(model_dir, only_table=True, csv_path=None):
+
+    eval_data = GetAAEvaluations(model_dir)
+
+    # get the total number of chi1 values
+    total_num_chi1 = 0
+    total_num_chi2 = 0
+    total_num_residues = 0
+    total_correct_num_chi1 = 0
+    total_correct_num_chi2 = 0
+    for key in eval_data.keys():
+        total_num_chi1 += eval_data[key].num_valid_chi1
+        total_num_chi2 += eval_data[key].num_valid_chi2
+        total_num_residues += eval_data[key].num_residues
+        total_correct_num_chi1 += eval_data[key].num_correct_chi1
+        total_correct_num_chi2 += eval_data[key].num_correct_chi2
+
+    if not only_table:
+        print('total num valid chi1 angles: ', total_num_chi1)
+        print('fraction correct:', float(total_correct_num_chi1) / total_num_chi1)
+        print('total num valid chi2 angles: ', total_num_chi2)
+        print('fraction correct:', float(total_correct_num_chi2) / total_num_chi2)
+        print('total num residues: ', total_num_residues)
+
+        for key in eval_data.keys():
+            print(key)
+            if eval_data[key].num_valid_chi1 > 0:
+                print('chi1 accuracy: ', 
+                      eval_data[key].num_correct_chi1/eval_data[key].num_valid_chi1, 
+                      eval_data[key].num_valid_chi1, eval_data[key].num_residues)
+            if eval_data[key].num_valid_chi2 > 0:
+                print('chi2 accuracy: ',
+                      eval_data[key].num_correct_chi2/eval_data[key].num_valid_chi2,
+                      eval_data[key].num_valid_chi2, eval_data[key].num_residues)
+            if eval_data[key].num_valid_chi2_given_chi1 > 0:
+                print('chi2 accuracy given chi1: ',
+                      eval_data[key].num_correct_chi2_given_chi1/eval_data[key].num_valid_chi2_given_chi1,
+                      eval_data[key].num_valid_chi2_given_chi1)
+
+    # output for csv table
+    csv_out = list()
+    csv_out.append('AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD')
+    AA = ['ARG','ASN','ASP','CYS','GLN','GLU','HIS','ILE','LEU','LYS','MET','PHE','PRO','SER','THR','TRP','TYR','VAL']
+    for aa in AA:
+        data = eval_data[aa]
+        num = '%i'%(int(data.num_valid_chi1))
+        chi1_fraction = ' '
+        chi2_fraction = ' ' 
+        chi2_fraction_given_chi1 = ' '
+        rmsd = ' '
+        if data.num_valid_chi1 > 0:
+            chi1_fraction = '%.2f'%(100*float(data.num_correct_chi1)/data.num_valid_chi1)
+        if data.num_valid_chi2 > 0:
+            chi2_fraction = '%.2f'%(100*float(data.num_correct_chi2)/data.num_valid_chi2)
+        if(data.num_valid_chi2_given_chi1) > 0:
+            chi2_fraction_given_chi1 = '%.2f'%(100*data.num_correct_chi2_given_chi1/data.num_valid_chi2_given_chi1)
+        if len(data.rmsd_values) > 0:
+            rmsd = '%.2f'%(np.mean(data.rmsd_values))
+        csv_out.append(','.join([aa, num, chi1_fraction, chi2_fraction, chi2_fraction_given_chi1, rmsd]))
+  
+    if csv_path:
+        with open(csv_path, 'w') as fh:
+            fh.write('\n'.join(csv_out))

sidechain/do_plot.py

+import matplotlib.pyplot as plt
+import numpy as np
+import analyze
+
+model_dir_one = 'models/promod/frm'
+model_dir_two = 'models/scwrl/frm'
+
+label_one = 'ProMod3'
+label_two = 'SCWRL4'
+
+fig_out_path = 'chi1_accuracy_frm.png'
+
+one = analyze.GetAAEvaluations(model_dir_one)
+two = analyze.GetAAEvaluations(model_dir_two)
+
+one_chi1 = list()
+two_chi1 = list()
+AA = ['ARG','ASN','ASP','CYS','GLN','GLU','HIS','ILE','LEU','LYS','MET','PHE',
+      'PRO','SER','THR','TRP','TYR','VAL']
+for aa in AA:
+    one_chi1.append(one[aa].num_correct_chi1/one[aa].num_valid_chi1)
+    two_chi1.append(two[aa].num_correct_chi1/two[aa].num_valid_chi1)
+
+x_promod = list()
+x_scwrl = list()
+for i in range(18):
+    x_promod.append(i+0.1)
+    x_scwrl.append(i+0.5)
+
+cred = (128.0/255,0.0,0.0)
+cblue = (102.0/255,153.0/255,204.0/255)
+plt.bar(x_promod, one_chi1, width = 0.4, color=cred, label=label_one, 
+        linewidth=2.0, alpha=0.75, align='edge', edgecolor='k')
+plt.bar(x_scwrl, two_chi1, width = 0.4, color=cblue, label=label_two, 
+        linewidth=2.0, alpha=0.75, align='edge', edgecolor='k')
+plt.ylim((0.5, 1.0))
+plt.legend(frameon=False, loc='upper left')
+x_tick_positions = list()
+for item in x_promod:
+  x_tick_positions.append(item + 0.4)
+plt.xticks(x_tick_positions, AA, rotation=30)
+plt.savefig(fig_out_path)
+plt.clf()

sidechain/do_supplemental_tables.py

+import analyze
+
+model_directories = ['models/promod/frm',
+                     'models/promod/frm_no_subrotamer_optimization',
+                     'models/promod/rrm',
+                     'models/promod/bb_indep_frm',
+                     'models/promod/bb_indep_rrm',
+                     'models/scwrl/frm',
+                     'models/scwrl/rrm']
+
+
+# you'll get a different number of evaluated aspartic acids in this testset when
+# evaluating SCWRL4 and ProMod3
+# The bad guy is: 2pst, X.ASP9
+# The problem is: a backbone oxygen is missing.
+# SCWRL4 completely deletes the residue
+# ProMod3 doesn't touch it and the correct sidechain remains in place
+# This is unfair in favour of ProMod3, as the correct sidechain is evaluated.
+# however, the effect on overall performance can considered to be neglectible...
+
+for md in model_directories:
+    print('processing:', md)
+    csv_path = '_'.join(['eval', md.split('/')[-2], md.split('/')[-1] + '.csv']) 
+    analyze.PrintPerformance(md, False, csv_path=csv_path)
+

sidechain/eval_promod_bb_indep_frm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,68.09,68.20,74.18,2.02
+ASN,2875,76.17,39.76,47.49,0.75
+ASP,4013,71.72,53.35,62.09,1.07
+CYS,999,84.68, , ,0.22
+GLN,2501,72.57,62.69,72.62,1.16
+GLU,4611,65.80,62.18,69.28,1.59
+HIS,1542,79.64,45.72,49.59,1.03
+ILE,3964,91.60,84.06,86.97,0.34
+LEU,6554,85.66,83.25,94.14,0.38
+LYS,3819,68.29,73.21,76.88,1.23
+MET,1406,78.73,67.07,76.60,0.81
+PHE,2715,90.39,85.27,88.96,0.77
+PRO,3230,79.01,78.20,98.79,0.17
+SER,4101,59.33, , ,0.41
+THR,3784,75.29, , ,0.38
+TRP,979,83.55,71.50,81.05,1.26
+TYR,2346,89.00,81.50,86.11,0.90
+VAL,5018,87.23, , ,0.35
\ No newline at end of file

sidechain/eval_promod_bb_indep_rrm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,66.98,69.98,74.71,2.07
+ASN,2875,74.57,37.18,44.82,0.78
+ASP,4013,70.15,50.83,59.68,1.13
+CYS,999,84.38, , ,0.22
+GLN,2501,71.57,61.22,70.06,1.19
+GLU,4611,63.96,62.61,68.90,1.67
+HIS,1542,78.08,40.53,44.44,1.09
+ILE,3964,90.94,83.25,86.10,0.35
+LEU,6554,84.82,83.06,94.39,0.40
+LYS,3819,67.32,73.24,77.17,1.26
+MET,1406,77.52,67.00,77.52,0.84
+PHE,2715,86.37,79.52,85.12,0.91
+PRO,3230,78.30,77.43,98.73,0.17
+SER,4101,58.55, , ,0.41
+THR,3784,74.71, , ,0.38
+TRP,979,80.59,66.09,75.03,1.45
+TYR,2346,85.51,79.28,85.14,1.07
+VAL,5018,87.35, , ,0.35
\ No newline at end of file

sidechain/eval_promod_frm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,73.67,68.65,73.80,1.96
+ASN,2875,83.34,49.04,56.43,0.63
+ASP,4013,82.01,61.50,70.86,0.77
+CYS,999,87.89, , ,0.18
+GLN,2501,75.77,63.97,72.03,1.11
+GLU,4611,70.11,62.91,70.40,1.49
+HIS,1542,84.95,48.57,52.21,0.89
+ILE,3964,95.86,86.28,88.05,0.26
+LEU,6554,88.74,85.96,94.77,0.33
+LYS,3819,74.94,74.00,77.88,1.12
+MET,1406,81.58,72.05,80.38,0.74
+PHE,2715,94.00,87.62,90.56,0.61
+PRO,3230,80.93,80.00,98.81,0.14
+SER,4101,69.06, , ,0.32
+THR,3784,89.27, , ,0.21
+TRP,979,89.68,76.51,83.37,0.99
+TYR,2346,92.16,84.83,88.39,0.74
+VAL,5018,93.08, , ,0.25
\ No newline at end of file

sidechain/eval_promod_frm_no_subrotamer_optimization.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,73.76,71.26,76.05,1.97
+ASN,2875,83.17,49.15,56.71,0.63
+ASP,4013,82.28,61.48,70.32,0.77
+CYS,999,87.69, , ,0.18
+GLN,2501,76.33,63.89,71.50,1.11
+GLU,4611,70.07,65.26,72.79,1.48
+HIS,1542,85.02,48.44,51.79,0.89
+ILE,3964,95.71,86.15,87.98,0.26
+LEU,6554,88.59,85.92,94.64,0.33
+LYS,3819,75.18,74.18,77.88,1.13
+MET,1406,81.93,72.12,80.03,0.75
+PHE,2715,92.04,87.11,90.16,0.67
+PRO,3230,80.93,80.00,98.81,0.14
+SER,4101,69.03, , ,0.32
+THR,3784,89.19, , ,0.21
+TRP,979,87.54,75.69,82.03,1.07
+TYR,2346,90.58,84.78,88.33,0.80
+VAL,5018,93.20, , ,0.24
\ No newline at end of file

sidechain/eval_promod_rrm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,71.79,68.93,73.93,2.03
+ASN,2875,81.81,46.09,53.66,0.66
+ASP,4013,80.36,58.58,68.74,0.83
+CYS,999,87.99, , ,0.18
+GLN,2501,75.53,63.13,70.62,1.14
+GLU,4611,68.08,63.70,71.39,1.56
+HIS,1542,84.18,44.36,48.07,0.93
+ILE,3964,95.23,85.19,87.05,0.28
+LEU,6554,87.46,84.67,94.49,0.35
+LYS,3819,73.61,73.42,77.20,1.16
+MET,1406,81.01,70.70,79.10,0.78
+PHE,2715,91.42,82.80,86.62,0.70
+PRO,3230,80.25,79.29,98.77,0.14
+SER,4101,69.06, , ,0.32
+THR,3784,89.16, , ,0.21
+TRP,979,87.23,71.60,78.81,1.17
+TYR,2346,89.09,81.80,87.03,0.89
+VAL,5018,93.12, , ,0.25
\ No newline at end of file

sidechain/eval_scwrl_frm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,74.23,71.20,76.81,1.93
+ASN,2875,81.60,47.62,55.24,0.67
+ASP,4012,81.56,60.82,70.17,0.82
+CYS,999,87.99, , ,0.20
+GLN,2501,75.37,63.97,72.41,1.13
+GLU,4611,69.29,65.73,73.49,1.49
+HIS,1542,83.72,45.59,49.26,0.95
+ILE,3964,95.31,84.76,86.69,0.27
+LEU,6554,87.60,85.26,94.79,0.36
+LYS,3819,74.31,74.10,77.80,1.14
+MET,1406,80.94,72.33,79.88,0.78
+PHE,2715,92.08,85.56,87.96,0.72
+PRO,3230,80.80,79.41,98.20,0.14
+SER,4101,69.11, , ,0.33
+THR,3784,89.48, , ,0.21
+TRP,979,87.44,76.00,82.48,1.10
+TYR,2346,90.49,83.08,86.62,0.86
+VAL,5018,92.63, , ,0.26
\ No newline at end of file

sidechain/eval_scwrl_rrm.csv

+AA,num,X1 correct (%),X2 correct (%),X2 correct given X1 (%),avg RMSD
+ARG,3601,72.06,68.29,73.87,2.02
+ASN,2875,80.87,46.75,54.75,0.68
+ASP,4012,79.49,57.63,67.89,0.88
+CYS,999,87.59, , ,0.20
+GLN,2501,74.09,63.17,71.40,1.16
+GLU,4611,67.53,64.32,72.03,1.55
+HIS,1542,83.46,42.48,46.15,0.97
+ILE,3964,94.75,83.70,85.92,0.28
+LEU,6554,86.41,83.98,94.40,0.38
+LYS,3819,72.72,73.13,76.99,1.18
+MET,1406,79.23,70.77,78.82,0.83
+PHE,2715,90.87,82.10,85.37,0.77
+PRO,3230,79.72,78.27,98.10,0.15
+SER,4101,67.93, , ,0.34
+THR,3784,88.82, , ,0.22
+TRP,979,85.90,68.74,76.10,1.23
+TYR,2346,88.49,80.18,85.02,0.97
+VAL,5018,92.07, , ,0.27
\ No newline at end of file

sidechain/get_scwrl_testset_structures.py

+import os
+from ost.mol.alg import MolckSettings, Molck
+from ost import conop, io
+import traceback
+
+with open('scwrl_testset_ids.txt') as fh:
+    data = fh.readlines()
+testset_ids = list()
+for line in data:
+    testset_ids += line.strip().split()
+
+outdir = 'scwrl_testset'
+if not os.path.exists(outdir):
+    os.makedirs(outdir)
+
+compound_lib = conop.GetDefaultLib()
+molck_settings = MolckSettings(rm_unk_atoms=True,
+                               rm_non_std=True,
+                               rm_hyd_atoms=True,
+                               rm_oxt_atoms=True,
+                               rm_zero_occ_atoms=True,
+                               colored=False,
+                               map_nonstd_res=True,
+                               assign_elem=True)
+
+for item in testset_ids:
+    try:
+        print('processing ', item)
+        out_path = os.path.join(outdir, item + '.pdb')  
+        prot = io.LoadPDB(item, remote = True)
+        io.SavePDB(prot.Select('peptide=true'), out_path)
+        molcked_out_path = os.path.join(outdir, item + '-molcked.pdb')
+        Molck(prot, compound_lib, molck_settings)
+        io.SavePDB(prot.Select('peptide=true'), molcked_out_path)
+    except:
+        print('failed for', item)
+        traceback.print_exc()
+