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 273

results do not provide evidence of further cor-
related behaviour between these two residues.
However, the same authors proposed that the
stability at the N-terminus of β-strand C might
be enhanced by the relatively short turn connect-
ing β-strands B and C (turn BC). Our data show
that, although with different solvent exposure
profiles across the five MD trajectories, residues
Ala36 (β-strand B), Asp39 (turn BC) and Glu42
(β-strand C) are found to be in the same clusters,
showing that these residues tend to have the same
collective behaviour, and indicating that in fact
the residues at the end of β-strand B, turn BC,
and beginning of β-strand C could belong to the
same folding unit .

interactions in the pathways to form the molecular
intermediates responsible for protein aggregation
and amyloid formation. The identification of such
residues and interactions is vital for the detailed
understanding of the molecular mechanisms of
amyloid diseases and for the development of
rational approaches towards anti-amyloid drugs.
Moreover, the use of a physics-based simulation
approach and the identification of correlated be-
haviour in SASA variation patterns upon protein
unfolding across several protein structural classes,
may help in the future the development of new and
improved knowledge-based potentials for protein
unfolding, protein folding and protein structure
prediction studies.

concluSIon AcknoWledgMent

One of the challenges for data analysis in multiple
protein unfolding simulations is to identify, among
several physical and structural properties those that
are essential in describing the unfolding and folding
processes. Looking at a wide range of properties
and experimental conditions further increases the
potential amount of data generated by such simula-
tion models. Analyzing and interpreting these data
requires automated methods such as data mining.
Here, we showed the application of two data
mining methods for the analysis and comparison
of the solvent accessible surface area (SASA)
variation of each one of the amino-acid residues
in the amyloidogenic protein Transthyretin, across
multiple molecular dynamics protein unfolding
simulations. Some of the identified residues have
been previously recognized by experimental and
computational approaches as important in the
unfolding process of TTR and may shed some new
and helpful insights in the understanding of TTR
amyloidosis. The ability to find and characterize
clusters of amino-acid residues with correlated sol-
vent exposure behaviour upon protein unfolding
in amyloidogenic and non-amyloidogenic proteins
may help identify critical residues and critical

The authors acknowledge the support of the
“Fundação para a Ciência e Tecnologia”, Portugal,
and the program FEDER, through grant PTDC/
BIA-PRO/72838/2006 (to PJA and RMMB) and
the Fellowships SFRH/BD/16888/2004 (to CGS)
and SFRH/BPD/42003/2007 (to PGF). We thank
the Center for Computational Physics, Physics
Department, University of Coimbra and Computer
Science and Technology Center, Informatics De-
partment, University of Minho, Braga, Portugal,
for the computer resources provided for the MD
simulations.

referenceS

Adcock, S. A., & McCammon, J. A. (2006). Mo-
lecular dynamics: Survey of methods for simulat-
ing the activity of proteins. Chemical Reviews ,
106 (5), 1589-1615. doi:10.1021/cr040426m

Agrawal, R., & Srikant, R. (1994). Fast algorithms
for mining association rules. In 20th International
Conference on Very Large Databases (pp. 487-
499). San Francisco: Morgan Kaufmann.