Visualization of normal modes from an elastic network model

The elastic network model (ENM) is a fast method to estimate movements in a protein structure, without the need to run time-consuming MD simulations. A protein is modelled as mass-and-spring model, with the masses being the \(C_\alpha\) atoms and the springs being the non-covalent bonds between adjacent residues. Via normal mode analysis distinct movements/oscillations can be extracted from the model.

An anisotropic network model (ANM), is an ENM that includes directional information. Hence, the normal mode analysis yields eigenvectors, where each atom is represented by three vector components (x, y, z). Thus these vectors can be used for 3D representation.

In the case of this example a normal mode analysis on an ANM was already conducted. This script merely takes the structure and obtained eigenvectors to add a smooth oscillation of the chosen normal mode to the structure. The newly created structure has multiple models, where each model depicts a different time in the oscillation period. Then the multi-model structure can be used to create a video of the oscillation using a molecular visualization program.

The file containing the eigenvectors can be downloaded via this link.

normal modes
# Code source: Patrick Kunzmann
# License: BSD 3 clause

from tempfile import NamedTemporaryFile
from os.path import join
import numpy as np
from numpy import newaxis
import biotite.structure as struc
import as strucio
import as mmtf
import biotite.database.rcsb as rcsb

# A CSV file containing the eigenvectors for the CA atoms
VECTOR_FILE = "../../download/glycosylase_anm_vectors.csv"
# The corresponding structure
# The normal mode to be visualized
# '-1' is the last (and most significant) one
MODE = -1
# The amount of frames (models) per oscillation
# The maximum oscillation amplitude for an atom
# (The length of the ANM's eigenvectors make only sense when compared
# relative to each other, the absolute values have no significance)

# Load structure
mmtf_file =, "mmtf"))
structure = mmtf.get_structure(mmtf_file, model=1)

# Filter first peptide chain
protein_chain = structure[
    & (structure.chain_id == structure.chain_id[0])
# Filter CA atoms
ca = protein_chain[protein_chain.atom_name == "CA"]

# Load eigenvectors for CA atoms
# The first axis indicates the mode,
# the second axis indicates the vector component
vectors = np.loadtxt(VECTOR_FILE, delimiter=",").transpose()
# Discard the last 6 modes, as these are movements of the entire system:
# A system with N atoms has only 3N - 6 degrees of freedom
#                                   ^^^
vectors = vectors[:-6]
# Extract vectors for given mode and reshape to (n,3) array
mode_vectors = vectors[MODE].reshape((-1, 3))
# Rescale, so that the largest vector has the length 'MAX_AMPLITUDE'
vector_lenghts = np.sqrt(np.sum(mode_vectors**2, axis=-1))
scale = MAX_AMPLITUDE / np.max(vector_lenghts)
mode_vectors *= scale

# Stepwise application of eigenvectors as smooth sine oscillation
time = np.linspace(0, 2*np.pi, FRAMES, endpoint=False)
deviation = np.sin(time)[:, newaxis, newaxis] * mode_vectors

# Apply oscillation of CA atom to all atoms in the corresponding residue
oscillation = np.zeros((FRAMES, len(protein_chain), 3))
residue_starts = struc.get_residue_starts(
    # The last array element will be the length of the atom array,
    # i.e. no valid index
for i in range(len(residue_starts) -1):
    res_start = residue_starts[i]
    res_stop = residue_starts[i+1]
    oscillation[:, res_start:res_stop, :] \
        = protein_chain.coord[res_start:res_stop, :] + deviation[:, i:i+1, :]

# An atom array stack containing all frames
oscillating_structure = struc.from_template(protein_chain, oscillation)
# Save as PDB for rendering a video with PyMOL
temp = NamedTemporaryFile(suffix=".pdb")
strucio.save_structure(, oscillating_structure)
# biotite_static_image = normal_modes.gif


Gallery generated by Sphinx-Gallery