Computational Intelligence - November 2012 - 59

For positive data points, the final Positive Score (PS) is
calculated as
PS = 50 + normalized score # 50.
For negative data points, the final Negative Score (NS) is
calculated as
NS = 50 - normalized score # 50.
The final scores are between 0 and 100. For a positive vector, it will be more than 50, and for a negative one, it will be
less than 50.
To compare this score of each data point with GDT_TS
score given in CASP, Pearson's correlation is calculated for
both scores. In statistics, the Pearson product-moment correlation coefficient (sometimes referred to as the PPMCC
or PCCs, and typically denoted by r) is a measure of the
correlation (linear dependence) between two variables X
and Y, giving a value between +1 and −1 inclusive. The
correlation coefficient ranges from −1 to 1. A value of 1
implies that a linear equation describes the relationship
between X and Y perfectly, with all data points lying on a
line for which Y increases as X increases. A value of −1
implies that all data points lie on a line for which Y decreases as X increases. A value of 0 implies that there is no linear
correlation between the variables. Suppose we have two
variables X and Y, with means X and Y respectively and
standard deviations v x and v y respectively. The correlation
is computed as
r=

/ in= 1 (X i - X ) (Yi - Y )
(n - 1) v x v y

This correlation is calculated for every method submitted to
CASP per target. In following tables the correlation between
the few selected techniques from CASP 9 and GDT_TS score
is given along with EE_IFDT's correlation with GDT_TS.
From the tables, the enhanced encoding scheme making
use of the improved fuzzy decision trees has the lowest
Table 5 Pearson's correlation for target T0635: EE_IFDT
compared with other CASP competitors.

METhoD

ToTAl
TEMPlATES
nuMbEr oF wITh AvAIlAblE PEArSon'S
TEMPlATES rESulTS
CorrElATIon

QMEANclust

387

324

MulticomCluster

387

correlation with GDT_TS. EE_IFDT also provides results
for all available models whereas others provide results only
for a small percentage. The scoring technique used with
EE_IFDT certainly needs more refinement as it does not
consider parameters unique to model prediction techniques.
Even though EE_IFDT is not a good classifier or evaluator
when it comes to CASP templates, we still have to look at
the novel methodology used. EE_IFDT does not use
parameters from other scoring techniques. It uses only the
coordinate information from the 3D structure itself to make
the decision. Even though the EE_IFDT performance is
subpar in above tables, in case of Target T0635 it does perform better than some of submitted methods in CASP like
(Baltymas, Splicer, Splice_QA, PconsR, PconsD etc.). These
results are shown in Table 7.
In this paper, we use EE_IFDT to examine the templates
submitted to recent CASP competitions. First, we selected
only templates that are classified as either good or bad and
used them as test data set to evaluate (around 10% of all
templates in CASP 8 and 9 since we only chose model 1 of
all groups). The prediction accuracy using CASP template is
around 70%, (as shown in Table 1) in testing phase I. To get
a better understanding and to compare EE_IFDT with
other methods, two targets from CASP 9 were selected for
further investigation. The templates were classified as positive or negative based on their GDT_TS scores (Tables 3
and 4). In testing phases I and II, no scoring scheme was
used, since the only way to compare EE_IFDT with other
CASP competitors is to score each model. A rudimentary
scoring scheme was introduced, and its performance evaluated in comparison to other prominent CASP competitors
(Tables 5 and 6). EE_IFDT has less correlation scores compared with others. It was also noted that EE_IFDT scored
all models whereas other approaches scored a subsample of
templates. CASP competitors are not required to score every
model, so it is difficult to make any comparison. In Table 7,
EE_IFDT is better compared with some other competitors
of CASP.
Table 6 Pearson's correlation for target T0578: EE_IFDT
compared with other CASP competitors.

METhoD

ToTAl
nuMbEr oF
TEMPlATES

TEMPlATES
wITh AvAIlAblE
rESulTS

PEArSon'S
CorrElATIon

0.998

QMEANclust

625

321

0.814

325

0.998

MulticomCluster

625

322

0.794

Distill_NNPIF 387

325

0.937

Distill_NNPIF

625

322

0.498

ProQ2

387

314

0.728

ProQ2

625

314

0.572

EE_IFDT

387

0.678

EE_IFDT

625

0.050

november 2012 | Ieee ComputatIonal IntellIgenCe magazIne

Table of Contents for the Digital Edition of Computational Intelligence - November 2012