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 271

Table 3. Selected association rules mined from the SASA data derived from WT-TTR MD unfolding
simulations. These rules were selected from the subset of rules common to the largest number of MD
simulations. All the amino-acid residues listed in each rule are hydrophobic and have SASA value between
0% and 25%. The Support value (%) for each rule in a particular data set and the lowest Confidence
value (%) obtained for the association rule in the data sets are reported. Labelled with × are rules not
identified in a particular data set.

Association Rule Support Confidence

Run 1 Run 2 Run 3 Run 4 Run 5

AR1 Leu12 ← Ala25 & Ile73 & Phe95 30.709 × 47.082 51.456 × 92.966

AR2 Leu12 ← Phe33 & Ile107 & Ala109 48.819 × 37.995 42.632 × 90.127

× ×

AR3 Leu12 ← Pro11 & Val14 & Val28 51.419 56.455 37.591 98.236

AR4 Leu12 ← Val16 & Val32 & Leu58 48.569 × × 52.281 64.721 97.828

AR5 Trp79 ← Val32 & Ala91 & Ala108 32.271 × 70.754 53.093 × 90.663

AR6 Ile107 ← Ala25 & Ile73 & Leu111 34.521 × 50.531 53.093 × 93.892

AR7 Leu111 ← Leu12 & Phe64 & Ala109 44.569 × × 71.766 30.305 94.040

Although in Table 3 only 7 out of a total 98 rules
co-occurring in three of the sets of association rules
are shown, it generally describes the structure of
the rules obtained. We found out that only a small
subset of the hydrophobic residues appears in the
consequent member of the rules: Leu12 (57 rules),
Trp79 (17 rules), Ile107 (23 rules) and Leu111
(1 rule). The distribution of these residues in the
native structure of WT-TTR is shown in Figure 4.
Furthermore, we also observed that (i) all 98 rules
can be derived for data sets Run 1 and Run 4, (ii)
none of the 98 rules can be derived for data set
Run 2, and (iii) if one rule is derived for data set
Run 3, then it was not derived for data set Run 5,
and vice-versa. Driven by these observations, we
analysed the residues involved in these association
rules, and their SASA behaviour.
First, for the amino-acid residues involved
in the 98 rules co-occurring in three of the sets
of association rules two major SASA variation
profiles are observed: (i) the amino-acid residues
maintain the same values of solvent exposure
throughout the entire simulations; or (ii) at some
point of the simulation the amino-acid residues
move rapidly from positions of low exposure (≤
25%) to positions of high exposure ( ~ 100%).

This is the most frequent SASA variation profile
observed. Furthermore, these rules are describing
mainly events in the beginning of the simulation
when the conformations are more closely related
to the native structure of WT-TTR. Thus, for most
rules what we observe is the point in time in the
unfolding process that an event occurs which
drives the rupture of some structures, mainly in
β-strands A, B, F and G, causing the residues to
move to positions of high exposure to the solvent.
The data also shows that a group of residues as-
sociated with Trp79 maintains positions of low
exposure to the solvent till late in Run 3 and Run 4
( ~ 6th ns), but in Run 1 they move to positions of
high exposure much earlier, around the 3rd ns.
Second, particular subsets of rules occur dis-
jointly in different data sets. This seems to indi-
cate a degree of similarity/dissimilarity between
different simulations. In a sense, it is shown that
specific subsets of hydrophobic residues undergo
different unfolding routes: particular groups of
hydrophobic residues relate in a similar manner
in some simulations, but not in others. Based on
the 98 association rules found to occur in three
data sets, we identified the group of residues
responsible for the differences observed: Pro11,