clustering_of_pdb_models.py

## \example em2d/clustering_of_pdb_models.py
# This example clusters pdb models of an structure, chosen from a
# selection file.
#
# It is assumed that all the pdb files belong to the same structure
# and that the order of the atoms in the pdb files is the same in all files.
#
# After the clustering procedure, a linkage matrix is generated.
#

import IMP
import IMP.algebra
import IMP.core
import IMP.atom
import IMP.em2d
import sys
import csv

IMP.setup_from_argv(sys.argv, "clustering of PDB models")

"""
    Clustering of pdb models.
    This script clusters pdb models of an structure, chosen from a
    selection file.
    - It is assumed that all the pdb files belong to the same structure
    and that the order of the atoms in the pdb files is the same in all files
    - After the clustering procedure, a linkage matrix is generated.


"""

if sys.platform == 'win32':
    sys.stderr.write("this example does not currently work on Windows\n")
    sys.exit(0)


def get_columns(fn, cols=[], delimiter=" ", comment="#"):
    """ ge the columns of a file:
        cols - a list of columns to extract. E.g [0,3,5]
               If empty, all the columns are extracted
        lines starting with the comment character are ignored """
    columns = [[] for i in cols]
    # get a reader for the file
    reader = csv.reader(
        open(fn, "r"), delimiter=delimiter, skipinitialspace=True)
    for row in reader:
        if(row != [] and row[0][0] != comment):  # not empty or comment row
            if(cols == []):
                for i in range(0, len(row)):
                    columns[i].append(row[i])
            else:
                for i in range(0, len(cols)):
                    columns[i].append(row[cols[i]])
    return columns


def argmin(sequence):
    """ Argmin function: Returns the pair (min_value,min_index),
        where min_index is the index of the minimum value
    """
    min_value = sequence[0]
    min_index = 0
    for i in range(0, len(sequence)):
        if(sequence[i] < min_value):
            min_value = sequence[i]
            min_index = i
    return min_value, min_index


fn_selection = IMP.em2d.get_example_path("all-models-1z5s.sel")
fn_em2d_scores = IMP.em2d.get_example_path("em2d_scores_for_clustering.data")
# Load models
print("Reading models ...")
model = IMP.Model()
ssel = IMP.atom.ATOMPDBSelector()
coords = []
fn_models = IMP.em2d.read_selection_file(fn_selection)
n_models = len(fn_models)
hierarchies = []
for fn in fn_models:
    fn_model = IMP.em2d.get_example_path(fn)
    h = IMP.atom.read_pdb(fn_model, model, ssel, True)
    hierarchies.append(h)
    xyz = IMP.core.XYZs(IMP.atom.get_leaves(h))
    coords.append([x.get_coordinates() for x in xyz])

print("Computing matrix of RMSD ...")
rmsds = [[0.0 for i in range(0, n_models)] for n in range(0, n_models)]
transformations = [[[] for i in range(0, n_models)]
                   for j in range(0, n_models)]
# fill rmsd and transformations
for i in range(0, n_models):
    for j in range(i + 1, n_models):
        if(i != j):
            t = IMP.algebra.get_transformation_aligning_first_to_second(
                coords[i],
                coords[j])
            transformations[i][j] = t
            transformations[j][i] = t.get_inverse()
            temp = [t.get_transformed(v) for v in coords[i]]
            rmsd = IMP.algebra.get_rmsd(temp, coords[j])
            rmsds[i][j] = rmsd
            rmsds[j][i] = rmsd

# cluster
print("Clustering (Complete linkage method)...")
cluster_set = IMP.em2d.do_hierarchical_clustering_complete_linkage(rmsds)
mat2 = cluster_set.get_linkage_matrix()
print("Complete Linkage Matrix")
for m in mat2:
    print(m)

# Read scores from the scores file
em2d_scores = get_columns(fn_em2d_scores, [1])
em2d_scores = em2d_scores[0]

# get biggest clusters below a certain rmsd
rmsd_cutoff = 1.4
print("clusters below cutoff", rmsd_cutoff, "Angstroms")
clusters = cluster_set.get_clusters_below_cutoff(rmsd_cutoff)
for c in clusters:
    elems = cluster_set.get_cluster_elements(c)
    scores_elements = []
    for cid in elems:
        scores_elements.append(em2d_scores[cid])
    print("Cluster", c, ":", elems, scores_elements, end=' ')
    # find model with best score
    min_value, min_index = argmin(scores_elements)
    min_elem_id = elems[min_index]
    # The representative element is the one with the minimum em2d score
    print("representative element", min_elem_id, min_value)
    for i in elems:
        pdb_name = "cluster-%03d-elem-%03d.pdb" % (c, i)

        if(i != min_elem_id):
            print("Writing element", i, "aligned to ", min_elem_id, ":",
                  pdb_name)
            T = IMP.core.Transform(transformations[i][min_elem_id])
            ps = IMP.atom.get_leaves(hierarchies[i])
            for p in ps:
                T.apply_index(model, p.get_particle_index())
        else:
            print("Writing representative element", min_elem_id, ":", pdb_name)
        IMP.atom.write_pdb(hierarchies[i], pdb_name)