Using Data Mining Techniques to Probe the Role of Hydrophobic Residues in Protein Folding and Unfolding Simulations - Evolving Application Domains of Data Warehousing and Mining - page 262

in this work, with well-validated empirical poten-
tials implemented in the CHARMM27 forcefield
(MacKerell, 1998). Simulations were performed at
500 K in the NVE (constant number of particles,
constant volume and constant energy) ensemble,
using periodic boundary conditions. The length of
the simulations for this study was 10 ns using a
time step of 2 fs. Coordinates for the whole system
were saved every 1 ps. Short-range non-bonded
interactions were calculated with a 12 Å cut-off
with the pair list distances evaluated every 10
steps. Long-range electrostatic interactions were
treated using the particle mesh Ewald summa-
tion algorithm and were computed at every step.
The system details and other relevant simulation
parameters can be found elsewhere (Rodrigues
& Brito, 2004; Rodrigues, 2009).

tended conformation. SASA was computed with
the program NACCESS (Hubbard, 1993) using
a spherical probe of 1.4 Å radius, mimicking a
water molecule.

MethodS

clustering

Cluster analysis is an exploratory technique which
might be applied to group a collection of objects
into subsets or clusters, such that objects within
each cluster are more closely related to one another
than objects assigned to different clusters. Cluster
analysis has proven to be a valuable instrument in
life sciences, namely in genomics (Boutros, 2005;
Zhao, 2005) and proteomics (Meunier, 2007), al-
lowing to interpret changes in the expression of
entire groups of genes and to discover functional
relationships among them. For the particular
problem of protein folding or unfolding, it is par-
ticularly interesting to understand how amino-acid
residues relate to each other during the process.
Indeed, to discover groups of residues that change
solvent exposure in a coordinated fashion across
several unfolding simulations might be crucial to
define the folding nuclei of a protein (Brito, 2004;
Hammarström & Carlsson, 2000).
Clustering algorithms belong to the class of
unsupervised methods, i.e. they do not require
prior classification of the training data. They are
mainly used for pattern discovery, providing the
identification of novel and unexpected patterns
in the data set. From this class of algorithms,
agglomerative hierarchical clustering is one
of the most commonly used. The agglomera-
tive hierarchical clustering algorithm works by
successively grouping the most similar pairs of
objects. The algorithm starts by comparing each
pair of objects, then selects the two objects with
the most similar characteristics, groups these
together into a node, and repeats the procedure
with the remaining objects. This process contin-

Md Simulation Analysis

The microscopic data obtained from MD simula-
tions, such as the atomic positions and velocities,
can be used to calculate macroscopic properties
following changes in the structure of the protein
being simulated (Adcock & McCammon, 2006).
Some of the time-dependent properties commonly
monitored are the radius of gyration, the root
mean square deviation, the number of hydrogen
bonds, and the native contacts. Here, the solvent
accessible surface area (SASA) of each individual
amino-acid residue of the protein TTR along the
MD unfolding simulations was calculated in order
to study potentially correlated behaviour among
different amino-acid residues.
The solvent accessible surface area (SASA) of
a protein is defined as the locus of the centre of
a probe sphere (representing a solvent molecule)
as it rolls over the surface of the protein (Lee &
Richards, 1971). The relative SASA of a residue
reflects the percentage of the surface area of the
residue that is accessible to the solvent. It is defined
as the ratio between the SASA of the residue (X)
in the three-dimensional structure of the protein
and its SASA in a tripeptide (Ala-X-Ala) in ex-