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 269

Association Rules Extraction their initial SASA value. On the second half of the

simulation, the accessible surface increases and
begins to vary around a SASA value close to the
one in the native structure of TTR.
Panels G to I of Figure 3 depict the variation
of the solvent accessible surface area of residues
Leu12, Val71 and Ile107, respectively. The SASA
profiles shown are representative of the different
profiles observed for these residues, across the five
data sets. In simulation 1 (Panel G), the residues
are buried during most of the simulation, but in the
last nanoseconds become progressively exposed.
In Runs 2, 3 and 5 (Panel H), the residues are
buried in the interior of the protein throughout
the simulation, displaying a very small accessible
surface to the solvent, even late in the simulations
when the protein is already denatured. Finally, in
Run 4, the residues show low solvent accessibility
until the ~ 6 th ns of the simulation and then get
highly exposed to the solvent (Panel I).

Association rules were extracted using CAREN
(Azevedo, 2003), an association rule engine with
several rule derivation and selection features. As
we were interested in sets of association rules
characterising the relations between hydrophobic
amino-acid residues with SASA ≤ 25% involv-
ing at least 4 residues, we used the features for
antecedent and consequent filtering, and for mini-
mum number of items (residues) in a rule. The
χ 2 test during itemsets mining was also applied,
significantly reducing the number of relevant
itemsets. The standard confidence metric was
used to evaluate the association rules, and it was
set to 90%. The minimum support was set to 30%.
In Table 2, the association rules derived from the
five data sets (Run 1 to Run 5) are described in
terms of number of rules generated, total number
of residues involved in the rules, and support and
confidence intervals.
Across the five data sets, there is a large varia-
tion in the number of association rules generated,
but the number of amino-acid residues identified
is similar. Additionally, the support and confi-
dence values vary within comparable intervals.
As expected, support values for association rules
involving more residues tend to be lower, and as-
sociation rules with high support involve fewer
residues.

Association rules

Association rules were extracted from five data
sets (Run 1 to Run 5) describing the solvent ac-
cessible surface area (SASA) variation profiles
of each one of the 127 amino-acid residues that
constitute the monomer of WT-TTR throughout
five independent MD unfolding simulations. All
data sets were discretised to evaluate three levels
of solvent exposure: low ([0, 25]), medium (]25,
75[) and high ([75, 100[).

Table 2. Characterisation of the association rules involving hydrophobic residues, extracted from the
SASA variation profiles in five independent MD unfolding simulations of WT-TTR

Run 1 Run 2 Run 3 Run 4 Run 5

No. rules 10831 407 3876 32086 758

No. residues 28 28 27 29 24

Support (%) ]30, 84[ ]30, 74[ ]30, 89[ ]30, 85[ ]30, 88[

Confidence (%) [90, 100] [90, 100[ [90, 100] [90, 100] [90, 100]