Co
m
pu
ter Sci
ence a
nd Inf
or
mat
i
on
Tec
h
no
lo
gies
Vo
l.
1
, No
.
3
,
Novem
ber
2020
, p
p.
106
~
11
5
IS
S
N:
27
22
-
3221
,
DOI: 10
.11
591
/
csi
t.v
1i
3
.p
106
-
11
5
106
Journ
al h
om
e
page
:
http:
//
ia
esprime
.com/i
ndex.
php/csit
Genome
featur
e op
timizati
on and
corona
ry ar
tery di
sease
predicti
on
using cu
ckoo se
arch
E. N
eel
im
a
1
, M.S
. Pr
asad
Babu
2
1
Depa
rtment of
Com
pute
r
Scie
n
ce
& Engi
ne
eri
n
g
,
GITAM
Univ
ersity
,
Visakha
p
at
nam,
Andhra
P
rad
esh,
India
2
Depa
rtment of
CS
&SE,
Andhra
Univer
sit
y
,
Visa
khapa
tn
am,
And
hra
Prade
sh
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n 2
3
, 2
0
2
0
Re
vised
Ju
n
11
, 20
2
0
Accepte
d
J
un
2
8
, 20
2
0
Cardi
ovasc
u
la
r
d
isea
ses
(
CVD
)
is
among
th
e
m
aj
or
health
ail
m
ent
issue
le
ad
ing
to
m
il
l
i
ons
of
de
at
hs
e
ver
y
y
e
ar
.
In
r
ec
en
t
p
ast,
an
aly
z
ing
g
ene
expr
ession
da
ta
,
par
ticula
r
l
y
usin
g
m
ac
hine
learni
ng
strat
eg
ie
s
to
pre
dict
and
cl
assif
y
th
e
given
unl
abele
d
gen
e
expr
ession
re
c
ord
is
a
g
ene
r
o
us
rese
arch
issue.
Conce
rn
in
g
thi
s,
a
subs
ta
n
tial
req
u
ire
m
ent
is
fea
tur
e
opti
m
i
za
t
ion,
whic
h
is
since
th
e
ove
ral
l
gen
es
obser
ved
in
hum
an
b
od
y
ar
e
cl
osel
y
25000
and
among
the
m
63
6
are
c
ard
io
v
asc
ula
r
relat
ed
ge
nes.
Henc
e,
it
c
om
ple
xes
th
e
proc
ess
of
tr
ai
n
i
ng
the
m
a
chi
ne
learni
ng
m
odels
using
the
se
e
nti
re
ca
rd
i
o
vasc
ula
r
g
ene
f
ea
tur
es.
T
h
is
m
anusc
ript
uses
b
idi
re
ct
ion
al
pool
ed
var
i
ance
strat
eg
y
of
AN
OV
A
standa
rd
t
o
sele
c
t
opt
imal
fea
tur
es.
Along
the
sid
e
to
surpass
the
con
strai
nt
observe
d
in
tr
adi
t
ional
c
la
ss
ifi
ers,
whi
ch
is
unstable
ac
cur
acy
at
k
-
fo
ld
cro
ss
va
li
da
tion,
th
is
m
anusc
r
ipt
proposed
a
class
ifi
cation
strat
eg
y
that
bu
il
d
upon
the
sw
arm
int
ellige
n
ce
te
chn
ique
c
a
ll
ed
cuc
ko
o
sea
rch
.
Th
e
exp
e
rimenta
l
stud
y
in
dic
a
ti
ng
th
at
the
num
ber
of
opti
m
al
fe
at
ur
es
those
sel
ec
t
ed
b
y
propos
ed
m
odel
is
subs
ta
ntially
low
th
at
compare
d
to
t
h
e
othe
r
contem
pora
r
y
m
odel
tha
t
se
le
c
ts
fe
at
ur
es
usi
ng
forward
fe
at
u
re
se
lecti
o
n
and
class
ifi
es
u
sing
SV
M
cl
assifie
r
(FF
S&S
VM
).
The
expe
r
i
m
ent
al
stud
y
evi
nc
ed
th
at
the
proposed
m
odel
,
which
sel
ects
fe
at
ure
b
y
bidi
r
ecti
onal
pool
ed
var
ia
n
ce
esti
m
ation
and
cl
assifi
e
s
using
proposed
cl
assifi
cation
strat
eg
y
that
buil
d
on
cuc
koo
sea
rch
(BPV
E&CS
) outperforme
d
th
e s
e
lecte
d
co
nte
m
pora
r
y
m
odel
(FF
S&S
V
M).
Ke
yw
or
d
s
:
Cl
assifi
cat
ion
Com
pu
te
r
intel
li
gen
ce
Corona
ry arte
r
y disease
Cuck
oo sea
rch
Gen
e
exp
ressio
n
ca
d gen
e
s
Pr
e
dicti
ve
anal
ysi
s
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
E.
Neelim
a
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce & E
ng
i
nee
rin
g,
GI
T
AM
Un
i
ve
rsity
, V
isa
kh
a
pa
tnam
, A
ndhra
Pr
a
des
h,
I
nd
ia
.
Em
a
il
:
eadh
a.ne
el
i
m
a@g
m
ai
l.
com
1.
INTROD
U
CTION
Am
on
g
th
e
va
r
iou
s
healt
h
as
pe
ct
s
that
le
ad
to
deat
hs
,
ca
r
dio
va
scular
dise
ases
(
CV
D
)
ar
e
on
e
of
the
m
ajo
r
fact
or
s
that
le
ad
to
m
i
ll
ion
s
of
deat
hs
gl
ob
al
ly
eve
ry
ye
ar
[1
]
.
A
cute
MI
(My
oc
ard
ia
l
in
far
ct
i
on)
i
s
resu
lt
ant
el
em
ent
of
t
he
m
yocard
ia
l
ti
ssu
e
f
orm
ation
becaus
e
of
re
duced
bl
ood
s
upply
t
o
t
he
hear
t
a
nd
it
causes
resu
lt
s
in
m
i
ll
i
on
s
of
deaths
[1]
.
Ma
ny
sci
entifi
c
stu
dies
hav
e
f
ocused
on
so
l
utions
in
te
rm
s
of
diagnosi
s
,
pr
e
ve
ntion,
a
nd
c
ur
e
f
or
M
I,
bu
t
sti
ll
the
optim
al
su
ccess
not
accom
plishe
d
in
te
rm
s
of
m
it
igati
ng
the
m
or
ta
li
ty
rati
o
le
d
res
ult
ing
du
e
t
o
MI
issues.
In
t
he
pr
ese
nt
sce
nar
i
o,
pr
e
do
m
inan
tl
y
clinical
sy
m
pto
m
s
are
us
ed
f
or
diag
nosis
of
MI.
Ce
rtai
n
sym
pto
m
s
li
ke
com
plexiti
es
in
breat
hing,
i
ncon
ven
ie
nce
or
uneasi
ness
face
d
by
the
patie
nts
li
ke
ch
est
pain
,
repo
rts
of
a
bnorm
al
Ele
ct
ro
Ca
rd
i
o
Gr
am
(ECG
)
re
su
lt
s,
a
bnorm
al
fall
in
the
ci
rc
ul
a
ti
on
le
vels
of
c
Tns
(
card
ia
c
t
rop
on
i
ns
)
[
2].
T
ho
ugh
t
her
e
are
m
any
de
velo
pm
ent
s
that
has
ta
ke
n
place
i
n
t
he
do
m
ai
n,
sti
ll
there
are
c
ertai
n
li
m
it
at
io
ns
a
nd
c
on
st
raints
face
d
i
n
at
ta
ining
accu
rate
analy
sis
usi
ng
the
cu
rr
e
nt
dia
gnos
ti
c
syst
e
m
s.
Fo
r
i
ns
ta
nce
,
the
c
on
te
m
po
ra
ry
m
et
ho
ds
a
nd
s
olu
ti
ons
that
wer
e
pro
posed
in
hs
-
c
T
n
as
say
s
ha
s
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Geno
me
fe
at
ure
opti
miza
ti
on
and
c
oro
na
ry
ar
te
ry
disease
pr
e
dicti
on
us
in
g
c
uckoo
searc
h
… (
E. Neel
im
a
)
107
resu
lt
ed
i
n
im
pro
ved
sc
ope
of
detect
ing
t
he
lowe
r
ci
rc
ulati
ng
T
n
co
nc
entrati
ons
(
with
inc
rease
d
se
ns
it
ivit
y
towa
rd
s
analy
s
is).
Howe
ver,
on
e
of
t
he
key
co
ns
trai
nts
f
r
om
the
pro
ces
s
is
the
rise
of
false
al
arm
rates
as
a
gr
eat
er
num
ber
of
no
n
-
disease
d
pe
op
le
are
al
so
sho
wn
as
pron
e
to
c
onditi
ons,
becau
se
of
change
re
su
lt
in
g
i
n
cTns
due
to
the
oth
e
r
c
om
plication
(this
ref
le
ct
s
reduce
d
se
nsi
ti
vity
)
[3
]
.
T
he
oth
e
r
diag
nos
ti
c
m
et
ho
d
u
se
d
f
or
detect
ion
is
the
card
ia
c
m
i
RNAs
that
c
on
si
de
red
as
sensiti
ve
bi
om
ark
ers
[4
]
,
bu
t
few
li
m
it
a
ti
on
s
li
ke
the
lo
w
abun
dan
ce
,
ti
ssu
e
sp
eci
fic
e
xpressio
n
issues
a
nd
the
sm
al
l
siz
e
ha
s
im
paired
t
he
reli
abili
ty
ove
r
t
he
m
od
e
l.
T
he
ro
le
li
ke
the
bio
m
ark
ers
has
beco
m
e
m
or
e
sign
ific
a
nt
bec
ause
of
in
ve
ntion
of
fast,
im
p
rove
d,
a
nd
a
ut
om
at
ed
detect
ion
syst
e
m
s
[5
]
.
I
n
som
e
of
the
oth
e
r
stu
dies
that
carried
out
in
the
li
nes
of
de
fining
bio
m
ark
ers
f
or
diag
nosis,
CR
P,
B
NP
a
nd
oth
er
su
c
h
ki
nd
of
inflam
m
at
or
y
m
ark
ers
to
o
consi
der
e
d,
howev
e
r
only
m
a
rg
i
nal
i
m
pr
ovem
ents in
the ac
cu
racy l
evels
wer
e att
ai
ned
a
s the
ou
tc
om
e [6
-
8].
Do
m
ai
n
kn
ow
l
edg
e
of
the
pa
tho
lo
gical
a
nd
physi
ol
og
ic
al
as
pects
the
ke
y
aspects
reli
ed
upon
f
or
dev
el
op
i
ng
m
a
ny
of
the
ea
rlie
r
car
diac
bi
oma
rk
e
rs.
Wherea
s,
the
m
ic
ro
arra
y
platfo
rm
s
c
on
si
der
t
he
e
xpressi
on
of
la
r
ge
num
ber
of
gen
es
in
si
m
ultaneou
s
m
ann
er
,
t
hat
f
ocuses
on
e
na
bling
gen
e
e
xpressio
n
pro
fili
ng
ac
r
os
s
var
ie
d
path
wa
ys
in
sim
ultan
eo
us.
T
he
af
or
esai
d
m
et
ho
d
has
the
cap
abili
ty
to
ind
i
cat
e
broa
d
ra
ng
e
of
path
ophysiol
ogic
al
processes
of
CV
D
in
m
or
e
ec
onom
ic
and
e
ff
ic
ie
nt
m
ann
e
r
[
9].
Ge
ne
e
xpressio
n
prof
il
in
g
exten
ds
dee
pe
r
tha
n
the
bi
oma
rk
e
rs
to
ide
ntify
m
or
e
pote
nt
ia
l
bio
m
ark
ers
that
earli
er
repor
te
d t
o
be
ass
ociat
ed
with
C
VD.
G
ene
e
xpres
sio
ns
us
ually
en
able
us
to
ide
ntify
a
nd
discov
e
r
insig
htf
ul
an
d
m
or
e
se
ns
it
ive
bio
m
ark
ers
tha
t
can
r
eflect
up
on
CV
D.
Ma
jo
rity
of
t
he
stu
di
es
that
ha
ve
f
ocused
i
n
t
his
sect
ion
has
pro
vid
e
d
sign
ific
a
nt
resul
ts
from
the
process.
I
n
[
10]
,
a
stu
dy
ca
rr
ie
d
ou
t
f
or
gen
e
e
xpressi
on
a
naly
sis
to
unde
rsta
nd
a
nd
disco
ver
co
nte
m
po
rar
y
an
d
s
ensiti
ve
bi
om
a
rk
e
rs
of
C
V
D
identifie
d
48
2
gen
es
that
a
re
in
as
so
ci
at
ion
to
com
po
sit
ion
of
coro
nar
y
at
heroscler
otic
plaq
ues
a
nd
m
ajo
rity
of
t
hem
nev
e
r
ta
gged
to
t
he
at
heroscler
os
is
[10].
In
[1
1],
wide
scal
e
gen
e
ex
pr
essi
on
pro
fili
ng
c
om
pr
isi
ng
56
div
e
rg
e
nt
genes
f
or
at
he
ro
scl
e
ro
ti
c
an
d
no
n
-
at
heroscler
otic
hu
m
an
co
r
on
a
r
y
arterie
s
e
xp
l
ored
,
of
w
hich
49
of
them
wer
e
associat
ed
wit
h
C
AD
earli
er
[11].
In
[
12]
,
t
he
aut
hors
ha
ve
f
oc
use
d
on
ide
ntifyi
ng
a
set
of
cl
a
ssifyi
ng
gen
e
s
base
d
on
dem
og
ra
phic
s
a
nd
it
ha
s
strongly
de
picte
d
t
he
obstr
uc
ti
ve
CA
D
in
non
-
diabeti
c
patie
nts
[
12]
.
Dive
rg
e
nt
ra
ng
e
of
ge
ne
e
xpres
s
ions
identifie
d
that
dif
fer
e
ntiat
ed
the
isc
hem
ic
an
d
non
-
isc
hem
ic
card
iom
yopathy
c
ondi
ti
on
s
of
the
pa
ti
ents
confro
nting
e
nd
-
sta
ges
[
13
-
14]
.
I
n
[
15
]
,
the
auth
or
s
ha
ve
w
orke
d
on
m
ic
r
oarray
analy
sis
and
ge
ne
ex
pressi
on
prof
il
in
g
t
hat
a
re
us
e
d
f
or
discov
e
rin
g
ge
nes
relat
ed
to
heart
fail
ures
base
d
on
e
xpres
sio
n
pro
file
s
of
pa
ti
ents
with
hear
t
fail
ure
com
plica
ti
on
s.
I
n
[
16]
,
the
stud
y
has
ta
r
ge
te
d
on
norm
al
con
t
ro
ls
a
nd
M
I
patie
nts
ha
ve
fou
nd
that
the
ge
neti
c
m
ark
et
s
an
d
the
der
e
gula
te
d
path
ways
tha
t
are
ass
ociat
ed
with
the
dise
ase
rec
urre
nce
in
first
tim
e MI p
at
ie
nt
s [
16]
.
It
is
i
m
per
at
i
ve
that
the
ef
fic
acy
with
w
hich
the
bloo
d
tran
scriptase
denotes
the
changes
of
transc
riptio
nal
el
e
m
ents
in
he
art,
im
pr
ov
es
t
he
accu
racy
of
diag
nosis.
I
n
[
17
]
,
t
he
aut
hor
s
hav
e
re
porte
d
that
upon
co
nducti
ng
a
ge
no
m
e
wide
s
urvey
by
us
ing
m
ic
roarr
ay
s
an
d
the
expresse
d
se
qu
e
nc
e
ta
gs
ha
ving
the
per
ip
he
ral
blood t
ra
ns
cri
pt
m
e to the
tra
nsc
ript m
e o
f
nin
e
oth
e
r h
um
a
n
ti
ss
ues
i
nclu
ding t
he one
s
of
hear
t,
m
or
e
than
80
%
of
overla
pp
i
ng
is
est
im
at
ed
at
ti
ssu
e
le
vels
.
84%
of
overla
pping
with
hear
t,
ind
ic
at
in
g
t
hat
study
of
pe
rip
her
al
b
lood
tra
ns
c
ript
ase
can
be
an
e
conom
ic
and
r
eadil
y
acce
ssib
le
too
l
for
pro
xy
gen
e
ex
press
ion
i
n
oth
e
r
ti
ssu
es
[
17
]
.
T
houg
h
m
any
stud
ie
s
hav
e
f
ocu
se
d
in
the
dom
ai
n
of
diff
e
re
ntial
expressio
n
i
n
CV
D
ou
tc
om
es,
in
[
18
]
,
t
he
a
uthor
s
ha
ve
f
oc
us
e
d
on
usi
ng
diff
e
re
ntial
ex
pr
e
ssion
f
or
cl
as
sifyi
ng
the
patie
nt
r
eco
r
d
ou
tc
om
es.
Su
c
h
a
n
ap
proac
h
pr
ov
i
des
e
ff
i
cacy
to
im
pr
ove
t
he
diag
no
sis
to
s
ub
-
cl
as
sify
patie
nts
.
Also,
the
disc
rim
inatory
featur
e
s
f
or
dif
fer
e
ntiat
ing
over
norm
al
pr
of
il
es
a
nd
t
he
pa
ti
ents
with
MI
,
CA
D
an
d
t
he
on
e
s
com
pr
isi
ng
un
sta
ble
an
gin
a
ov
e
r
ge
ne
e
xp
ressio
n
i
n
blood
cel
ls.
Bl
oo
d
tra
ns
c
riptase
that
us
e
d
with
easi
ly
acce
ssible
ti
ss
ue
f
or
the
dia
gnos
ti
c
pur
pose
s
an
d
m
ajo
rity
of
s
uch
c
on
t
rib
utions
de
pict
that
the
c
om
pu
ta
ti
on
al
ov
e
r
head
res
ulti
ng
f
r
om
den
s
e
nu
m
ber
of
ge
ne
feat
ur
es
are
adap
te
d
in
t
he
le
arn
in
g
proce
ss.
I
n
this
pa
pe
r,
a
t
-
te
st
dep
e
nde
nt f
eat
ur
e
o
ptim
izati
on
m
od
el
p
r
opos
e
d
w
hich
co
ul
d
e
ff
ect
ive
ly
reduce
t
he
c
ount o
f
fe
at
ur
e
s
use
d
for
analy
sis.
T
he
s
olu
ti
on
use
s
le
sser
num
ber
of
featu
res
when
co
m
par
e
d
to
m
any
of
the
ea
rlie
r
m
od
el
s.
D
espite
of
us
i
ng
li
m
it
e
d
set
of
featu
r
es,
the
acc
ura
cy
le
vels
of
di
agnosis
with
r
edu
ce
d
false
a
la
rm
rates
has
bee
n
the outcom
e for
the
prop
os
e
d solutio
n.
2.
RELATE
D
W
ORK
S
In
[
19
]
,
t
he
st
ud
y
has
detect
ed
var
ie
d
iss
ue
s
of
im
balances
that
m
igh
t
cree
p
up
i
n
t
he
us
a
ge
of
m
ic
ro
arr
ay
,
be
cause
of
noisy
,
huge
volum
e
and
i
rr
el
e
van
t
sam
ples.
Be
ca
us
e
of
t
he
af
ore
-
sta
te
d
c
om
pl
exiti
es,
researc
hers
f
oc
us
e
d
to
us
e
swa
rm
intel
l
igence
te
chn
i
qu
e
s
f
or
addressi
ng
t
he
issue.
T
he
stu
dy
us
ed
the
te
c
hniq
ue
of
ACO
sam
pl
ing
,
w
hich
de
velo
ped
base
d
on
A
nt
Colo
ny
Op
ti
m
iz
a
ti
on
(
ACO
)
al
go
r
it
h
m
fo
r
el
im
i
nating
the
noisy
a
nd
i
rr
el
eva
nt
featu
res
in
the
proc
ess
of
feature
s
el
ect
ion
.
S
VM
cl
assifi
ers
were
ada
pt
ed
beca
us
e
of
it
s
prom
inence
f
or
high
dim
e
ns
io
nal
data
cl
assifi
cat
ion
e
ve
n
with
sm
al
l
set
of
sam
ples.
The
issue
s
of
unsta
ble
cl
assifi
cat
ion
pe
rfor
m
ance
ide
ntifie
d
i
n
cr
os
s
validat
io
n
pro
cess
are
a
m
ajo
r
facto
r.
I
n
[
20]
,
the
a
utho
rs
ha
ve
adap
te
d
a
hy
br
i
d
m
o
del
f
or
sel
ect
in
g
op
ti
m
al
feature
s
by
us
in
g
Ar
ti
fici
al
Be
e
Col
on
y
(
AB
C)
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3, N
ov
em
ber
20
20:
106
–
11
5
108
the
cl
assifi
cat
ion
car
ried
out
us
in
g
t
he
S
VM
cl
assifi
ers.
AB
C
us
e
d
for
cl
us
te
ring
an
d
s
el
ect
ing
op
ti
m
al
featur
es,
wh
ic
h
re
duces
the
searc
h
s
pac
e.
Ex
per
im
ental
stud
ie
s
de
pict
that
the
unsta
ble
accu
racy
at
the
le
vel
of
10
-
f
old
cl
assifi
cat
ion
.
I
n
[21],
the
m
od
el
pr
opos
es
th
e
us
a
ge
of
AC
O,
a
nd
R
oughe
st
t
heory
(
RS
T
)
in
com
bin
at
ion
f
or
achievin
g
the
op
ti
m
iz
ed
featur
e
co
unt.
Acc
ur
acy
of
feature
sel
ect
ion
is
i
nv
e
rsely
propo
rtion
at
e
t
o
t
he
le
vel
of
dim
ension
al
it
y
in
the
f
eat
ure
set
.
In
[22],
it
e
xplo
re
s
the
us
a
ge
of
BA
T
al
gorithm
fo
r
reduci
ng
the d
im
ension
a
li
ty
o
f
feat
ur
e a
nd selec
ti
on of
op
ti
m
al
f
eat
ur
e
s.
In
[23],
f
uzzy
based
m
od
el
dep
ic
ti
ng
the
ru
le
s
dep
e
nd
ing
on
relat
io
ns
hi
p
am
on
g
the
f
eat
ures
dev
el
op
e
d,
us
i
ng
t
he
c
om
bin
at
ion
of
ACO
an
d
BA
T
te
ch
nique.
In
ad
diti
on,
the
r
ules
ca
n
be
in
use
f
or
sel
ect
in
g
op
ti
m
al
feature
s
in
dynam
ic
m
ann
er
.
Am
on
g
the
co
ns
trai
nt
s
that
en
visag
ed
in
the
m
odel
,
there
is
nee
d
f
or
expos
ur
e
to
en
su
re
sel
ect
ion
of
p
rio
r
at
trib
ut
es
that
s
uppor
ts
in
sel
ect
in
g
the
dep
e
ndent
at
tribu
te
s,
base
d
on
dev
ise
d
fu
zzy
r
ules.
RST
an
d
BC
O
com
bin
e
d
in
[
24
]
wh
e
re
in,
the
em
ph
asi
s
is
on
cl
us
te
ring
the
featu
res
base
d
on
phen
otype
or
the
patte
rn
that
ide
ntifie
s
the
optim
al
featur
es.
It
u
se
d
the
local
it
y
s
ensiti
ve
discri
m
inant
analy
sis
(
LS
D
A
)
for
re
duci
ng
dim
ension
al
i
ty
of
featu
re
se
ts,
wh
ic
h
f
ur
t
he
r
cl
us
te
rs
,
the
ou
tc
om
e
us
ing
f
uzzy
c
-
m
eans
(
FCM
)
al
go
rithm
.
FCM
us
e
d
in
c
om
bin
at
ion
wit
h
ABC
a
pproa
ch
for
featu
re
si
m
il
arity
assessm
en
t
wh
il
st
f
or
m
ing
the
cl
us
te
rs.
T
he
FCM
inc
orporated
with
a
rtific
ia
l
bee
co
lon
y
(
ABC
)
a
ppr
oac
h
that
use
d
f
or
featur
e
sim
il
ari
ty
assessm
ent
durin
g
cl
ust
er
f
or
m
at
ion
.
Ot
he
r
co
ntem
po
rar
y
m
od
el
s
in
the
featur
e
opti
m
izati
on
are
bin
a
ry
bat
al
gorithm
(
BBA
)
a
nd
ABC
wer
e
us
ed
[
25
]
,
a
nd
in
[
26
]
m
ini
m
u
m
red
unda
ncy
an
d
m
axim
u
m
releva
nce
(
M
-
RMR
),
a
nd
PS
O a
nd
DT
i
n
[
27]
.
The
M
-
RM
R
[
26
]
is
a
n
e
ffec
ti
ve m
et
ho
d
for
re
du
ct
io
n
of
noise
and irrele
va
nt fea
tures
a
pa
rt fr
om
r
edu
ci
ng th
e d
im
ension
al
it
y.
In
order
to
s
urpass
t
he
c
ons
trai
nts
obser
ve
d
i
n
e
xisti
ng
m
et
a
-
heu
risti
c
swa
rm
intel
lig
ence
-
based
featur
e
sel
ect
ion
m
od
el
s,
a
c
ouple
of
featu
r
e
sel
ect
ion
te
c
hn
i
qu
e
s
cal
le
d
f
orwa
rd
feat
ure
sel
ect
io
n,
f
orwa
r
d
featur
e
i
nclusi
on,
a
nd
bac
kw
ard
f
eat
ur
e
el
i
m
inati
on
discu
ssed
in
[
28]
.
T
he
ex
per
im
ent
al
stu
dy
i
nd
ic
at
ing
that,
a
m
on
g
these
t
hree
strat
e
gies
f
orward
featu
re
sel
ect
ion
is
op
tim
a
l.
Howe
ve
r,
the
pe
rfo
rm
a
nce
ob
se
r
ved
in
10
-
fo
l
d
cl
assifi
cat
ion
done
by
SVM
,
m
axi
m
u
m
c
la
ssific
at
ion
ac
cur
acy
li
m
it
ed
to
89
%
a
nd
no
t
co
ns
ist
ent
bet
we
e
n
div
e
rg
e
nt
f
old
s
.
The
cl
assifi
er
s
dep
ic
te
d
ab
ove
ha
ve
var
ie
d
le
vels
of
perf
or
m
ance
eff
ic
a
cy
that
influ
en
ced
by
pre
-
pr
ocessin
g
sta
ges
f
or
dataset
s.
P
re
-
proc
essing
sta
ge
s
de
picte
d
i
n
feat
ur
e
sel
ect
ion
proces
s
c
ou
l
d
le
ad
t
o
bette
r
perf
or
m
ances
f
or
cl
ass
ifie
rs.
Feat
ures
re
duct
ion
in
t
he
dataset
s
is
on
e
of
the
crit
i
cal
aspects
fac
ing
the
cl
assifi
er.
Ma
ny
of
t
he
ea
rlie
r
te
ch
niques
of
feat
ur
e
sel
ect
ion
or
re
du
ct
i
on
has
de
picte
d
that
it
c
ou
l
d
be
a
resou
rcef
ul
so
l
ution
f
or
cl
ass
ific
at
ion
purpo
ses.
I
n
a
ddit
ion
,
t
he
acc
ur
ac
y
and
pe
rfo
rm
ance
of
cl
assi
f
ic
at
ion
m
igh
t de
pe
nd
on the
qual
it
y of featu
re s
el
ec
ti
on
tech
ni
qu
es
ad
a
pted.
3.
CORO
NAR
Y AR
TE
RY D
I
S
EASE P
REDICTIO
N
F
ROM GE
NOME
FEATU
RES
US
I
NG
CUCK
OO
SE
ARCH
In
t
his
sect
io
n
of
stu
dy,
the
process
of
feat
ur
e
optim
iz
at
i
on
f
or
ge
no
m
e
featu
res
a
nd
i
n
te
rm
s
of
pr
e
dicti
ng
t
he
coro
nar
y
arte
r
y
disease
he
ur
ist
ic
scal
e
-
bas
ed
de
fini
ng
ba
sed
on
C
uc
koo
sear
ch
is
propose
d.
The
f
ur
t
her
sec
ti
on
s,
fi
rstly
th
e
m
et
ho
ds
an
d
m
at
er
ia
ls
us
ed
in
the
dev
ise
d
m
od
el
discusse
d.
F
ur
t
her
,
the
m
et
ho
d
of
fe
at
ur
e
optim
iz
ation
base
d
on
A
NOV
A
sta
nda
rd
te
r
m
ed
as
bi
dire
ct
ion
al
poole
d
va
riance
est
i
m
at
ion
discusse
d. I
n f
ur
t
her
a
nce,
t
he
searc
h process
and label
pred
ic
ti
on
b
a
sed
on
cu
c
koo sea
rc
h disc
u
sse
d.
3.1.
Metho
ds
an
d
mat
eri
als
3.1.1.
The fe
ature se
t
The
63
6
ge
nom
es
am
on
g
t
he
total
25000
genom
es
are
r
el
at
ed
to
the
CVD
[
29]
,
whic
h
is
usual
ly
dep
ic
te
d
as
C
A
D
ge
nes.
In
te
r
m
s
of
e
valuati
ng
th
e
c
orrelat
ion
am
ong
t
he
636
ge
nom
e
fe
at
ur
es,
high
le
vels
of
process
c
o
m
pl
exity
are
im
per
at
ive,
a
nd
it
ca
us
es
sig
nifican
t
ra
ng
e
of
false
al
arm
ing
over
t
he
predict
io
n
m
od
el
s.
Hen
ce
,
in
orde
r
to
en
sure
li
ne
r
an
d
lo
wer
l
evels
of
com
plexity
,
ens
ur
in
g
trun
cat
i
on
of
false
al
arm
rat
es
to
m
in
i
m
al
le
vels
is
ve
ry
esse
ntial
.
E
ver
y
rec
or
d
of
t
he
datase
t
ada
pted
f
or
t
r
ai
nin
g
a
nd
te
st
ing
phases
c
om
pr
ise
the
s
i
ng
le
nu
cl
eotide
poly
m
o
rphism
(
SNP
)
of
e
ver
y
ge
ne,
de
noti
ng
ge
ne
ti
c
var
ia
ti
on
of
var
i
ou
s
ge
ne
s.
In
add
it
io
n,
the
i
ni
ti
al
le
ng
th
of
e
ver
y
recor
d
is
636
value
s
depi
ct
ing
the
S
NPs
of
al
l
t
he
63
6
ge
no
m
es
that
li
ste
d
in
CA
D
Gen
e
s.
In
it
ia
l
dataset
com
pr
ise
s
the
s
et
of
rec
ords
t
ha
t
ei
ther
la
belle
d
as
pro
ne
t
o
C
VD
or
the
ones
that
are
sal
ubrio
us
with
no
t
race
of
a
ny
CV
D
im
pl
ic
at
ion
s.
I
n
ad
diti
on,
t
he
dim
ension
al
it
y
of
ge
nes
co
un
t
has
to
reduce
f
ro
m
th
e
c
urren
t
num
ber
of
636
to
c
onside
rab
ly
le
ss
er
value
s.
A
N
OVA
sta
ndar
d
te
rm
ed
as
bid
ir
ect
ion
al
pooled
var
ia
nc
e
est
im
a
ti
on
is
adap
te
d
f
or
the
process
of
re
duci
ng
the
dim
ension
al
it
y
to
opt
i
m
iz
e
the
ge
ne
count
and
buil
ding
t
he
pro
pose
d
s
cal
e.
I
n
a
dd
it
ion
,
t
he
detai
ls
of
bi
directi
on
al
po
oled
va
ri
ance
est
i
m
at
ion
t
hat
is
adap
te
d for
fea
ture o
pti
m
iz
ati
on pr
ocess
e
xp
lore
d
in
the
f
ollow
in
g sec
ti
on.
3.1.2.
Bi
dire
ctiona
l
pooled
va
ri
ance
est
im
at
io
n
Attrib
utes
of
e
ver
y
rec
ord
i
n
the
ch
os
e
n
dataset
denotes
eac
h
ge
ne
of
C
ADGen
e
s
f
or
t
he
c
ount
of
636.
Hen
ce
,
eve
ry
r
ecord
c
om
pr
ise
s
636
SN
P
s
a
s
values
pe
rtai
ning
to
al
l
the
genom
es.
To
de
fu
se
t
he
num
ber
of
gen
e
s
that
co
nsi
der
e
d
for
opt
i
m
al
featur
es,
the
c
ov
a
riance
am
idst
values
de
noti
ng
e
very
ge
ne
in
the
r
ecord
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Geno
me
fe
at
ure
opti
miza
ti
on
and
c
oro
na
ry
ar
te
ry
disease
pr
e
dicti
on
us
in
g
c
uckoo
searc
h
… (
E. Neel
im
a
)
109
la
belle
d
ei
the
r
as
pro
ne
or
s
al
ubrio
us
f
or
a
ll
the
fe
at
ur
e
s.
Ge
nes
are
op
t
i
m
al
featur
es
com
pr
isi
ng
ef
f
ect
ive
cov
a
riance
am
i
ds
t
value
s
pe
rtai
nin
g
t
o
pro
ne
or
the
sal
ubrio
us
rec
ords
c
ho
s
en.
F
or
est
im
at
i
ng
var
ia
nce
of
SN
Ps
,
com
pr
isi
ng
val
ues
of
a
ge
ne
r
el
at
ed
to
pr
on
e
or
s
al
ubrio
us
r
ecords
of
t
he
c
ho
s
en
t
rainin
g
set
,
the
m
et
ho
d
ada
pts
ANO
VA
sta
nd
ard
bid
irect
io
na
l
poole
d
va
ria
nce
e
stim
at
ion
.
Ba
sed
on
res
ults
en
visag
ed
in
[
30
-
31
]
,
the
m
et
ho
d
is
chosen
f
or
a
naly
sis.
T
he
bid
irect
io
nal
po
ol
ed
va
riance
es
tim
a
ti
on
is
a
da
pted
f
or
s
el
ect
i
ng
opti
m
a
l
featur
es
per
ta
ini
ng
to
e
ver
y
recor
d
(
bo
th
pro
ne
a
nd
sa
lubrio
us
)
f
or
a
trai
ning
set
c
hose
n.
Diff
e
re
ntial
values
am
idst
two
disti
nct v
ect
or
s
d
e
picte
d by th
e u
sa
ge of
b
i
directi
on
al
poole
d varia
nce esti
m
at
ion
as foll
ows:
=
(
⟨
1
⟩
−
⟨
2
⟩
)
√
(
1
)
+
(
2
)
In the e
quat
ion ab
ov
e
⟨
1
⟩
,
⟨
2
⟩
ind
ic
at
es
the
m
ean
values
i
de
ntifie
d
f
or
rel
evan
t
vecto
rs
1
,
2
and
t
hese
vector
s
i
nd
ic
at
e
th
e
SN
Ps
c
onsti
tuted
as
values
to
a
ge
ne
per
ta
i
ni
ng
to
rec
ords
la
belle
d
a
s
pro
ne
an
d
sal
ubri
ous
res
pecti
vely
in g
i
ven trai
ni
ng set.
The
re
pr
e
se
ntati
on
s
(
1
)
,
(
2
)
si
gnify
t
he
m
ean
s
qu
a
r
e
dista
nce
of
the
vecto
rs
1
,
2
resp
ect
i
vely
.
The
bid
i
recti
onal
poole
d
var
ia
nce
est
im
ation
is
the
rati
o am
i
ds
t
the
m
ean
va
riat
ion
of
r
el
at
ive
vect
ors
and
t
he
s
quare
root
of
s
um
of
m
ean
s
qu
a
re
di
sta
nces
of
t
he
r
el
at
ive
vect
or
s
.
I
n
f
ur
t
her
a
nce,
the
p
-
value
(
de
gr
e
e
of
pr
ob
a
bili
ty
)
[
32
]
is
at
ta
ined
base
d
on
t
-
t
able
[
33]
.
P
-
va
lue
is
m
uch
le
sser
t
han
the
pro
ba
bili
ty
thresh
ol
d,
wh
ic
h reflect
s
that t
he vect
ors
v
a
ry.
Hen
ce
, t
he feat
ur
e
den
oting res
pecti
ve
v
ect
ors a
re
of optim
al
f
eat
ure.
3.1.3.
Cu
ck
oo se
arc
h
The
nat
ur
al
el
em
ents
bas
ed
m
et
a
-
he
ur
ist
ic
s
m
od
el
s
de
velo
pe
d
a
re
am
ong
the
best
set
of
al
gorithm
s
to
address
the
iss
ues
of
op
ti
m
iz
a
ti
on
.
The
pr
opose
d
w
ork
e
valu
at
es
the
fitness
for
a
gi
ven
ge
ne
vect
or
f
or
co
r
on
a
ry
artery
disease
pro
ne
set
and
t
he
sal
ubri
ou
s
s
et
s
based
on
c
onte
m
po
rar
y
m
e
ta
-
he
ur
ist
ic
m
o
del
of
C
ucko
o
Searc
h
(CS)
[
34]
.
CS
dev
el
op
e
d
base
d
on
obli
gate
bro
od
par
a
sit
ism
of
the
c
ucko
o
sp
eci
es.
Its
m
ai
n
c
har
act
e
risti
c
is
to
le
t
the
eg
gs
in
the
nests
of
oth
er
bir
d
s
pecie
s
that
are
relat
ively
m
a
tc
hin
g.
Th
ree
key
f
undam
ental
s
based
on
su
c
h
nestin
g
process
f
ollo
we
d
by
Cuc
koo
are:
Cuc
koo
eg
g
de
no
te
s
a
s
olu
ti
on
t
o
the
iss
ue
a
nd
it
dr
op
s
ra
ndom
ly
in
a
c
hosen
nes
t.
H
ow
e
ve
r,
on
ly
on
e
e
gg
le
ft
at
ever
y
i
ns
ta
nc
e.
T
he
nests
t
hat
com
pr
ise
hi
gh
e
r
qu
al
it
y
of
e
gg
s
hav
e
t
o
pass
to
the
f
uture
ge
ne
rati
on
Nest
owne
r
sh
al
l
ide
nti
fy
a
cuc
koo
egg
base
d
on
pro
ba
bili
ty
∈
[
0,
1].
In
the
instance
of
su
ch
occ
urrenc
e,
the
ne
st
owner
le
a
ves
the
nest
an
d
de
vel
op
s
oth
e
r
nest
in
a
var
ie
d
loc
at
io
n.
The
c
um
ulati
ve
num
ber
of
ne
sts
is
the
fi
xed v
al
ue.
N
ot
al
l
the p
re
viously
m
entioned
r
ule
s
are
es
s
entia
l, as
the
cucko
o
searc
h
us
e
d
in
the
pro
po
s
ed
m
od
el
,
only
for
ide
ntifyi
ng
the
fitness
of
featu
res
f
or
a
cho
s
en
in
put
gene
record
.
He
nce,
the
pro
po
sal
is
to
dev
el
op
nest
s
in
a
t
rad
it
io
na
l
m
ann
er
a
nd
the
sea
rch
perf
or
m
ed
us
in
g
ra
ndo
m
appr
oach.
Tra
di
ti
on
al
searc
h
dro
ps
on
ly
one
egg
i
n
the
c
hos
en
ne
st,
but
in
the
pro
posed
s
olu
ti
on,
it
cl
ones
the
egg
t
o
var
ie
d
num
ber
of
com
patible
nests
an
d
places
one
e
gg
in
eve
ry
com
patible
nest.
It
al
so
est
im
at
es
fit
ness
of eve
ry eg
g f
or e
ntire n
e
st hi
erarc
hy.
3.1.4.
The d
ata
se
t
Data
set
ge
ne
ra
te
d
base
d
on
re
cords
de
noti
ng
coro
nar
y
a
rtery
s
us
cepti
bili
ty
m
od
e
(
NCBI
G
EO
Dataset
ID
:
G
DS4
527)
an
d
Atheroscl
ero
ti
c
C
oro
nary
A
rtery
Disea
se
pron
e
(N
C
BI
GE
O
Datas
et
I
D:
GDS36
90)
ar
e
gathe
red
f
ro
m
NCBI
gen
e
ex
pr
essi
on
om
nib
us
(
NCBI
GE
O)
[
35
]
,
a
uth
e
ntica
te
d
as
ge
ne
ex
pr
essi
on
da
ta
set
reposit
ory
.
T
he
dataset
G
DS452
7
c
om
pr
ise
ge
ne
ex
pressi
on
s
of
20
sub
je
ct
s.
Am
on
g
them
10
rec
ord
s
are
cat
egorized
as
sal
ubrio
us
a
nd
rest
of
the
rec
ords
are
cat
e
gor
iz
ed a
s
pro
ne t
o c
oronary
arte
ry
disease
. T
he
oth
e
r
dataset
G
DS
3690
c
om
pr
ise
s
153
rec
ords
of
w
hich
66
rec
ords
cat
eg
or
iz
e
d
as
sal
ubrio
us
and
rests
of
t
hem
as
pro
ne
to
c
oro
nar
y
a
rtery
dis
ease.
Ba
sed
on
the
rec
ords
of
t
wo
dataset
s
represe
ntin
g
173
s
ubje
ct
s,
values
ob
s
er
ved
for
CAD
Gen
e
s,
w
hich
are
c
ollec
te
d
as
recor
d
f
or
e
ver
y
s
ubj
ect
.
Stat
ist
ic
s
of
fi
nal
datase
ts
that
gen
e
rated
fro
m
the
process
d
e
picte
d
in
the
fo
ll
ow
in
g
T
a
ble
1
.
Table
1
.
Cl
assi
ficat
ion
of
da
ta
set
s
Data
Su
b
jects
Cu
m
u
lativ
e
record
s
173
Leng
th
of
eac
h
r
ec
o
rd
636
Reco
rds
with d
isea
se p
ron
e labelin
g
97
Reco
rds
of
salu
b
riou
s lab
elin
g
76
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3, N
ov
em
ber
20
20:
106
–
11
5
110
3.2.
Opt
im
iz
ing g
e
no
me
f
e
atures
As
a
pa
rt
of
portio
ning
proc
ess
of
la
belle
d
rec
ords
i
n
th
e
dataset
,
w
hich
cl
assifi
e
d
t
o
tw
o
s
et
s
ind
ic
ti
ng
co
r
onary
a
rtery
disease
pro
ne
a
nd
sal
ubrio
us
rec
ords
res
pecti
ve
ly
.
The
set
s
are
in
the
form
of
m
at
rix
siz
e
of
rec
ords
countin
g
as
row
co
unt
an
d
C
AD
ge
nes
c
ount
ed
as
col
um
n
count,
wh
ic
h
a
r
e
fixe
d
to
63
6
[
29]
.
E
ve
ry
r
ow
of
the
m
at
rix
sh
al
l
be
a
vect
or
de
no
ti
ng
SNPs
at
ta
in
ed
f
or
al
l
the
C
ADGe
nes
per
t
ai
nin
g
to
in
div
id
ual
c
ase
an
d
e
ver
y
colum
n
in
the
vecto
r
in
dicat
es
SNPs
gathered
f
r
om
sp
eci
fic
gen
e
i
n
the
chose
n
cases.
C
on
te
xt
of
opti
m
al
feat
ur
e
sel
ect
ion
is
ab
out
a
ge
ne
com
pr
isi
ng
a
va
ried
vect
or
of
S
NP
s
pe
rtai
nin
g
t
o
pro
ne
a
nd
sa
lu
br
i
ou
s
recor
d
s
et
s
.
In
a
dd
it
io
n,
it
a
pp
li
es
bi
directi
onal
pool
ed
var
ia
nce
es
tim
a
ti
on
te
st
ov
e
r
t
he
at
ta
in
ed value
f
or
a
gen
e
p
e
rtai
ning to
both
labell
ing
set
s usi
ng t
he follo
wing
proces
s.
ste
p 1:
∀
=
1
|
|
{
∃
∈
∧
∃
∈
}
Be
gin
ste
p 2:
⟨
⟩
=
∑
{
(
)
}
|
|
=
1
|
|
//
ob
ser
ving
t
he
m
ean
⟨
⟩
of
t
he
al
l
values
com
pr
ise
d
in
c
olum
n
vector
of
t
he
set
de
noti
ng S
NP
s
f
ound in
re
cords o
f
for g
ene
ste
p 3:
⟨
⟩
=
∑
{
(
)
}
|
|
=
1
|
|
//
obser
ving
t
he
m
ean
⟨
⟩
of
the
al
l
v
al
ue
s
c
om
pr
ise
d
i
n
col
um
n
vecto
r
of
the
s
et
that de
note
s S
NP
s
obser
ve
d
i
n
al
l rec
ords
of
the set
f
or g
e
ne
ste
p 4:
⇌
=
√
∑
(
)
−
⟨
⟩
|
|
=
1
|
|
−
1
+
∑
(
)
−
⟨
⟩
|
|
=
1
|
|
−
1
//
obse
rvi
ng
the
r
oo
t
m
ean
s
qua
re
distance
⇌
of
t
he vect
ors
an
d
ste
p 5:
⇌
=
(
⟨
⟩
−
⟨
⟩
)
⇌
//
Estim
a
ti
ng
the
bid
irect
io
na
l pool
var
ia
nc
e
score
of
t
he ve
ct
or
an
d
vect
or
com
par
ison
ste
p 6:
(
(
⇌
)
<
)
//
Up
on
insta
nce
of
de
gree
of
pro
bab
il
it
y
(
⇌
)
identifie
d
for
⇌
is
le
sser
tha
n
t
he pr
obabili
ty
thresh
old
(usuall
y 0.0
1,
0.0
5 or 0
.1)
giv
e
n
ste
p 7:
←
{
}
//
then
the
ℎ
gen
e
of
the
C
ADGen
e
s
set
is
de
li
ber
at
ed
as
optim
a
l
a
nd
m
oved
to
t
he
op
ti
m
al
g
ene s
et
ste
p 8:
En
d
3.3.
Cu
ck
oo Se
arc
h for
fitnes
s
asse
ssmen
t
This
sect
io
n
e
xp
l
or
es
the
pro
cess
of
fitness
assessm
ent
thr
ough
c
ucko
o
s
earch
.
T
he
ove
rall
proces
s
include
s
nest
f
or
m
at
ion
,
hier
arch
ic
al
sea
rc
h
to
noti
fy
the
f
it
ness
of
t
he
optim
al
featur
es
of
th
e
giv
e
n
r
ecord
towa
rd
s
pro
ne
to
Co
rona
ry
a
rtery
disease
a
nd
sal
ubrio
us
sta
te
.
Nest
f
or
m
at
ion
,
sea
rc
h
an
d
la
bel
pre
dicti
on
process
explo
r
ed
in
foll
owin
g sec
ti
on
s
3.4.
Nest
f
orm
at
i
on
In
orde
r
to
perform
the
cuckoo
searc
h,
t
he
hierar
c
hy
o
f
th
e
nests
s
hould
gen
e
rate
f
or
c
orres
pondi
ng
disease
pro
ne
and
sal
ubrio
us
set
s
,
.
The
op
ti
m
al
gen
e
feat
ures
re
pr
ese
nt
t
he
nests
in
a
hi
erarch
y
s
uch
t
hat
each
set
of
opti
m
al
gen
e
f
eat
ures
re
pr
ese
nts
a
un
i
qu
e
nest
i
n
hierar
c
hy
that
r
efer
red
f
ur
t
her
as
nest
represe
ntati
ve
set
.
T
he
opti
m
al
ge
ne
feat
ur
e
set
s
ex
plored
su
c
h
t
hat
eac
h
set
co
ntains
m
or
e
tha
n
one
ge
ne
featu
re
tha
t
are
highly
c
orrelat
e
in
reg
a
rd
t
o
the
t
heir
res
pe
ct
ive
S
NP
s
as
values
fou
nd
i
n
rec
ords
of
th
e
c
orrespo
nd
i
ng
set
s
,
.Furthe
r,
t
hese
nest
re
pr
es
enta
ti
ve se
ts
ref
e
rred
as
an
d l
et
the
hie
rar
c
hies
,
form
ed respect
ive
to
disease
pro
ne
and
s
al
ub
rio
us
set
s
,
us
in
g
t
he
se
ne
st
re
pr
e
se
ntati
ve
set
s
Further
,
t
he
un
i
que
value
set
s
{
1
,
2
,
.
.
,
}
as
eggs,
s
uch
that
each
eg
g
represents
t
he
values
of
a
gen
e
featu
res
in
nest
re
pr
e
s
entat
ive
set
{
∃
∈
}
an
d
exists
in
at
le
ast
on
e
reco
rd
of
the
r
especti
ve
rec
ords
-
set
,
s
houl
d
place
i
n
t
o
t
he
ne
st
represe
nted by
{
∃
∈
}
.
3.5.
Asses
sing
f
it
n
ess b
y
nes
t
se
ar
ch
The
fitness
of
the
giv
e
n
rec
ord
est
i
m
at
es
base
d
on
the
nu
m
ber
of
com
patible
nests
noti
ced
in
res
pecti
ve
hierar
c
hies
,
.
C
on
ce
r
ning
t
his,
f
or
eac
h
nest,
a
ny
e
gg
of
t
he
re
sp
ect
ive
ne
st
is
i
den
ti
cal
to
the
values
ob
s
er
ved
i
n
gi
ven
rec
ord
f
or
the
ge
ne
feat
ures
in
co
rr
es
pondin
g
ne
st
repre
sentat
ive
set
th
en
the
fitness
of
the
giv
e
n
record
in
relat
ed
t
o
c
orr
esp
onding
hier
arch
y
w
il
l
incr
e
m
ent
by
1.
T
hi
s
pr
act
ic
e
delivers
the
fitness
relat
ed
to
disease
pro
ne
an
d
sal
ubri
ous
sta
te
for
gi
ve
n
rec
ord
.
F
ur
t
he
r
the
fitness
ra
ti
o
of
the
giv
e
n
recor
d
ab
out
to
bot
h
hierar
c
hies
will
m
easur
e,
whic
h
is
the
a
verage
of
the
fitn
ess
relat
ed
to
nu
m
b
er
of
nes
ts
in
co
rr
es
po
nd
i
ng
hierar
c
hies.
T
he
n
t
he
root
m
ean
s
quare
d
ist
ance of
th
e fitne
ss v
al
ues
co
rr
e
sp
on
ding
t
o
bo
th h
ie
rar
c
hies
s
hould
m
easur
e.
The
n
these
fitness
r
at
ios
an
d
r
oo
t
m
ean
squa
re
distances
c
orres
pondin
g
t
o
both
hiera
rc
hies
will
use
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Geno
me
fe
at
ure
opti
miza
ti
on
and
c
oro
na
ry
ar
te
ry
disease
pr
e
dicti
on
us
in
g
c
uckoo
searc
h
… (
E. Neel
im
a
)
111
to
c
onf
irm
the stat
e
of
the
gi
ve
n
recor
d
is
pr
on
e
to
co
r
on
a
r
y
vasc
ular
dise
ase
or not
that
ex
pl
or
e
d
i
n
f
ollow
in
g
sect
ion
.
Th
e
m
at
hem
atical
m
o
del to a
ssess t
he
f
it
ness
foll
ows
ste
p 1:
Let
be
the
nest
represe
ntati
ve
set
s
(
see
sec
3.
C)
of
disea
se
pro
ne
a
nd
sa
lubrio
us
hierar
c
hies
,
respec
ti
vely
,
suc
h
t
hat
eac
h
ne
st
r
epr
ese
ntati
ve
s
et
co
ntains
a
s
et
of
hi
gh
ly
c
orrelat
ed
featur
e
s
ob
ta
in
ed fr
om
o
ptim
al
g
ene
f
eat
ur
e
s d
isc
ov
e
re
d
(
s
ee sec
3.
B
)
ste
p 2:
Let
be
t
he reco
rd co
ntains
SNPs res
pecti
ve
to
al
l o
ptim
al
f
e
at
ur
e
ge
nes
sel
ect
ed
(see
sec
3.
B)
ste
p 3:
Let
be
the
set
represe
nting
th
e
set
s
of
value
s
as
e
ggs
to
pl
ace
in
nests,
s
uch
that
eac
h
e
gg
co
ntain
s
the v
al
ues o
bs
e
rv
e
d
i
n
for
t
he
g
e
nes of res
pe
ct
ive n
est
re
pr
e
sentat
ive set
{
∃
∈
}
.
ste
p 4:
=
∑
{
1
∃
∈
∃
∈
∧
∈
}
|
|
=
1
//
add
1
t
o
dis
ease
pro
ne
fitn
ess
of
t
he
gi
ve
n
rec
ord
relat
ed
to
pr
one h
ie
ra
rch
y
if
egg
is com
patib
le
to plac
e i
n nest
in
pron
e
hierar
c
hy
.
ste
p 5:
⟨
⟩
=
|
|
//
Find
in
g
t
he f
it
ness
rati
o
⟨
⟩
relat
ed
to
disease
pro
ne hierarc
hy
ste
p 6:
=
∑
{
√
(
1
−
⟨
⟩
)
2
}
+
⟨
⟩
∗
(
|
|
−
)
|
|
=
1
|
|
//
T
he
fitnes
s
rati
o
discar
ds
f
r
om
1
for
nu
m
ber
of
ne
sts
com
patible
to
the
eg
gs
e
xist
i
n
a
nd
t
he
fitn
ess
rati
o
m
ultip
li
es
by
t
he
nu
m
ber
of
i
nco
m
patible
nests,
wh
ic
h
is
th
e
di
ff
e
ren
ce
bet
we
en
total
nu
m
ber
of
nests
a
nd
nu
m
ber
of
co
m
pat
ible
nests
that
de
note
d
a
s
|
|
−
ste
p 7:
=
∑
{
1
∃
∈
∃
∈
∧
∈
}
|
|
=
1
//
ad
d
1
to
sal
ubrio
us
fit
ness
relat
ed
t
o
sal
ubri
ous
hierar
c
hy
if e
gg
is com
patible
to plac
e in
n
es
t
in sal
ubrio
us hiera
rch
y
.
ste
p 8:
⟨
⟩
=
|
|
//
Find
in
g
t
he f
it
ness
rati
o
⟨
⟩
relat
ed
to
sal
ubri
ou
s
h
ie
rar
c
hy
ste
p 9:
=
∑
{
√
(
1
−
⟨
⟩
)
2
}
+
⟨
⟩
∗
(
|
|
−
)
=
1
|
|
//
f
ind
i
ng the
r
oo
t
m
ean s
qu
a
r
e d
ist
ance
of t
he
salu
br
io
us
ste
p 10
:
fitness
us
in
g
t
he
sim
il
ar p
r
oce
ss d
e
fine
d for
pro
ne
fitne
ss
r
m
sd
calc
ulati
on
in
step
6
3.6.
Disco
verin
g
t
he recor
d s
tat
e
The
fitness
rati
os
⟨
⟩
,
⟨
⟩
a
nd
r
oot
m
ean
s
quare
dist
ances
,
obta
ine
d
in
res
pecti
ve
to
disease
pro
ne
an
d
sal
ub
rio
us
hiera
rc
hies
,
fo
r
giv
e
n
i
nput
record
shou
l
d
us
e
to
la
bel
t
he
rec
ord
is
pro
ne
to
d
ise
as
e or salu
bri
ous.
Th
e
label s
houl
d
de
fine
u
si
ng the c
onditi
ona
l flo
w
that
fo
ll
ow
s:
ste
p 1:
(
⟨
⟩
≅
⟨
⟩
)
Be
gin
ste
p 2:
(
<
)
Be
gin
ste
p 3:
Label the
r
ec
ord
as
d
ise
ase
pr
on
e
ste
p 4:
En
d
//
of step
2
ste
p 5:
Else
(
>
)
Be
gin
ste
p 6:
Label the
r
ec
ord
as
salu
br
i
ou
s
ste
p 7:
En
d
//
of ste
p 5
ste
p 8:
Else
//
of
c
ondit
ion
i
n
ste
p 5
ste
p 9:
Re
cord
sta
te
is
am
big
uous
//
s
ince
the
fitnes
s
rati
os
an
d
r
oot
m
ean
sq
uare
distance
ob
ta
ined
f
or
bot
h
hierar
c
hies is
s
a
m
e
ste
p 10
:
En
d
//
of step
1
ste
p 11
:
Else
Begin
//
of c
onditi
on in
s
te
p
1
ste
p 12
:
(
⟨
⟩
>
⟨
⟩
)
Be
gin
ste
p 13
:
Label the
r
ec
ord
as
d
ise
ase
pr
on
e
ste
p 14
:
En
d
//
of step
11
ste
p 15
:
Else
(
⟨
⟩
<
⟨
⟩
)
Be
gin
ste
p 16
:
Label the
r
ec
ord
as
salu
br
i
ou
s
ste
p 17
:
En
d
//
of
ste
p 1
4
ste
p 18
:
Else
Begin//
of
conditi
on in
step
15
ste
p 19
:
Re
cord
sta
te
is
a
m
big
uous/
/
since
the
fitne
ss
rati
os
an
d
roo
t
m
ean
sq
ua
re
distance
ob
ta
in
ed
f
or
both
hierar
c
hies a
re
no
t m
eet
ing
th
e prescri
bed co
nd
it
io
ns
ste
p 20
:
En
d
//
of ste
p 1
8
ste
p 21
:
En
d
//
ste
p 11
3.7.
Empi
ri
cal
ana
lysis o
f th
e
pr
oposed
mo
del
The
e
xperim
e
ntal
stu
dy
co
nducted
on
da
ta
set
exp
l
or
e
d
in
sect
io
n
3.4).
I
n
orde
r
to
e
xp
l
ore
the
pe
rfor
m
ance
sign
ific
a
nce
of
t
he
pr
opos
e
d
m
od
el
that
in
corp
or
at
ed
the
featur
e
opti
m
izati
on
by
bid
i
re
ct
ion
al
pooled
var
ia
nc
e
an
d
cuc
koo
search
m
od
el
cl
assifi
er
(BP
VE&CS
),
the
exp
e
rim
ental
resu
lt
s
ob
ta
ine
d
a
nd
com
par
ed
to
th
e
oth
e
r
Co
ntem
porar
y
m
od
el
[
28
]
that
sel
ect
s
op
ti
m
al
featur
es
us
in
g
for
ward
sel
ect
ion
te
ch
ni
que
and cla
ssifie
s
usi
ng SV
M
clas
sifie
r
(
FFS
&S
VM).
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3, N
ov
em
ber
20
20:
106
–
11
5
112
The st
at
ist
ic
s
of
the
da
ta
set
use
d ca
n
de
pict
in
ta
ble
1 t
hat
exp
l
or
e
d i
n
3.4
)
. T
he
cl
assifi
ca
ti
on
pr
oces
s
on
sai
d
dataset
us
in
g
the
bo
t
h
m
od
el
s
do
ne
i
n
4
f
old
s.
In
a
dd
it
ion
,
the
perform
ance
assessm
ent
of
the
pro
po
s
ed
m
od
el
and
co
ntem
po
rar
y
m
od
el
dep
ic
te
d
us
in
g
cl
assifi
c
at
ion
assessm
ent
m
e
tric
s
[3
6]
su
ch
as
pr
ec
isi
on,
sensiti
vity
,
sp
e
ci
fici
ty
,
and
a
ccur
acy
.
The
r
esults
obta
ine
d
for
bo
t
h
the
pro
po
se
d
a
nd
con
te
m
po
ra
ry
m
od
el
dep
ic
te
d i
n
T
a
ble
2. T
he
nota
ti
on
us
e
d
as
r
ow a
nd co
l
um
n
head
e
rs
a
re:
PPV
:
Po
sit
ive
pr
e
dicti
ve valu
e (
or) precisi
on
TPR:
True P
osi
ti
ve
Ra
te
(
or)
Sens
it
ivit
y
TNR: Tr
ue Ne
gative Rat
e
(or
)
S
pecifici
ty
FN
R:
False
Ne
gative
rate (
or)
m
issi
ng
rate
FPR:
False P
osi
ti
ve
Ra
te
(
or)
fall
ou
t
ACC:
A
cc
ur
ac
y
The
pr
e
dicti
on
accuracy
of
t
he
bo
t
h
the
m
odel
s
ob
se
r
ved
from
the
exp
eri
m
ents
dep
ic
te
d
in
Fig
ur
e
1.
The
resu
lt
s
depi
ct
ed
in
Fig
ure
1
e
vin
ci
ng
that
the
cl
assifi
cat
ion
accu
racy
ob
serv
e
d
for
BP
VE
&CS
is
sta
bl
e
an
d
su
bst
antia
ll
y
hig
h
with
great
er
than
93%
that
com
par
ed
to
FFS
&SV
M
,
wh
ic
h
obser
ve
d
as
inc
on
sis
te
nt
and less t
ha
n 9
0%
.
Figure
1
.
The
c
la
ssific
at
ion
ac
cur
acy
rati
os
of BP
VE&CS
a
nd FFS&
S
VM
ob
s
er
ved f
ro
m
4
-
fo
l
d
cl
assifi
c
at
ion
Table
2
.
T
he
m
et
rics an
d
th
e
va
lues
ob
ta
ine
d from
4
-
f
old
cla
ssific
at
ion
us
in
g pro
po
se
d
a
nd
con
te
m
po
ra
ry
m
od
el
.
PPV
TPR
TNR
FNR
FPR
ACC
BPVE
&
CS
Fo
ld
#
1
0
.95
8
0
.95
8
0
.94
7
0
.04
2
0
.05
3
0
.95
3
Fo
ld
#
2
0
.92
3
1
0
.89
5
0
0
.10
5
0
.95
3
Fo
ld
#
3
1
0
.91
7
1
0
.08
3
0
0
.95
3
Fo
ld
#
4
0
.95
7
0
.91
7
0
.94
7
0
.08
3
0
.05
3
0
.93
FFS&S
VM
Fo
ld
#
1
0
.88
0
.91
7
0
.84
2
0
.08
3
0
.15
8
0
.88
4
Fo
ld
#
2
0
.87
5
0
.87
5
0
.84
2
0
.12
5
0
.15
8
0
.86
Fo
ld
#
3
0
.91
7
0
.91
7
0
.89
5
0
.08
3
0
.10
5
0
.90
7
Fo
ld
#
4
0
.84
0
.87
5
0
.78
9
0
.12
5
0
.21
1
0
.83
7
Figure
s
2
a
nd
3
dep
ic
ts
t
he
pe
rfor
m
ance
adv
a
ntage
of
t
he
BP
VE
&CS
over
F
FS&
S
VM
to
wa
rd
s
sensiti
vity
and
sp
eci
fici
ty
tho
se
re
fe
rs
the
sign
ific
a
nce
of
disease
sc
ope
pre
dicti
on
and
sig
nifican
ce
of
sal
ubrio
us
sta
te
pr
e
dicti
on
resp
ect
ively
.
The
pr
opose
d
m
od
el
cl
early
ou
t
perfor
m
ed
the
FFS
&SV
M
in
this
reg
a
rd.
The
Fi
gures
4
and
5
evi
ncin
g
the
false
ne
gat
ive
rate
or
m
is
sing
rate,
false
posit
ive
rate
or
fall
-
ou
t
t
hose
in
di
cat
es
predict
io
n
fail
ur
e
rate
of
disease
sc
ope
a
nd
sal
ub
riou
s
sta
te
re
spe
ct
ively
ob
se
r
ved
f
or
BPVE&C
S
a
nd
FFS
&S
VM.
From
the
de
picte
d
res
ults
of
pre
dicti
on
fail
ur
e
rate
f
or
disease
sco
pe
an
d
sal
ubrio
us
stat
e is m
uch
lo
w for
pro
po
se
d
m
od
el
t
hat c
om
par
ed
to FFS
&S
VM.
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Geno
me
fe
at
ure
opti
miza
ti
on
and
c
oro
na
ry
ar
te
ry
disease
pr
e
dicti
on
us
in
g
c
uckoo
searc
h
… (
E. Neel
im
a
)
113
Figure
2
.
The
s
ensiti
vity
(
dise
ase p
red
ic
ti
on
rate)
of
BPVE&C
S a
nd F
FS&
SV
M
obser
ve
d from
4
-
f
old
cl
assifi
cat
ion
Figure
3
.
The
s
pecifici
ty
(
sal
ubri
ou
s
stat
e
pr
e
dicti
on r
at
e)
of BP
VE&CS
and FF
S&S
V
M
ob
s
er
ved f
ro
m
4
-
fo
l
d
cl
assifi
c
at
ion
Figure
4
.
The
disease
pr
e
dicti
on
fail
ur
e
r
at
e
(f
al
se
neg
at
ive
r
at
e
) of B
PV
E
&
CS
and FF
S&S
V
M
ob
s
er
ved f
ro
m
4
-
fo
l
d
cl
assifi
c
at
ion
Figure
5
.
The
s
al
ubrio
us
stat
e
pr
e
dicti
on f
ai
lu
re r
at
e
(f
al
se
posit
ive
rate)
of BPV
E
&CS a
nd FF
S
&SV
M
ob
s
er
ved f
ro
m
4
-
fo
l
d
cl
assifi
c
at
ion
The
proces
s
c
om
plexity
ob
s
erv
e
d
i
n
both
trai
ning
a
nd
te
sti
ng
phases
de
picte
d
i
n
Fi
gures
6
a
nd
7
resp
ect
ively
.
F
ro
m
the
de
picte
d
fi
gures
,
it
is
obvi
ou
s
to
c
oncl
ude
that
pro
cess
com
pleti
on
ti
m
e
of
BPV
E&C
S
in
trai
ning
a
nd
te
sti
ng
phases
is
sign
ific
a
nt,
since
they
ar
e
m
uch
le
sser
than
t
he
pr
oc
ess
com
pleti
on
tim
e
ob
s
er
ved f
or F
FS&SVM
in
r
e
sp
ect
ive tr
ai
ning a
nd test
ing p
hases.
Figure
6
.
Proce
ss Com
pleti
on
tim
e o
f
trai
ni
ng
ph
a
se
ob
se
r
ved f
or
both B
PVE&C
S a
nd
FFS&SVM i
n 4
-
fo
l
d
cl
assifi
c
at
ion
Figure
7
.
Proce
ss Com
pleti
on
tim
e o
f
te
sti
ng
ph
a
se
ob
s
er
ved f
or both B
PV
E
&CS
and FFS&
SVM
in 4
-
fo
l
d
cl
assifi
cat
ion
4.
CONCL
US
I
O
N
Gen
e
e
xpressi
on
s
f
or
m
s
as
t
he
com
bin
at
io
n
of
m
any
genes
a
m
ong
the
thousa
nds
of
ge
nes
de
fin
e
d
un
ti
l
now.
Among
these
th
ou
san
ds
of
ge
nes
,
636
ge
nes
id
entifi
ed
as
car
dio
va
scular
re
la
te
d
that
are
usual
ly
ref
e
rs
as
CA
D
Gen
e
s
[
29
]
.
Sti
ll
this
cou
nt
of
ge
nes
is
h
ig
h
dim
ension
to
app
ly
m
achine
-
le
arn
i
ng
m
et
ho
ds
t
o
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2722
-
3221
Com
pu
t. Sci.
I
nf. Tec
hnol.
,
V
ol.
1
, N
o.
3, N
ov
em
ber
20
20:
106
–
11
5
114
le
arn
car
dio
va
scular
r
el
at
ed
inf
or
m
at
ion
.
C
on
ce
r
ning
this,
reducin
g
the
dim
ension
al
it
y
of
the
C
ADG
enes
is
essenti
al
facto
r
to
im
pr
ov
e
th
e
perform
ance
of
the
m
achine
le
arn
i
ng
proces
s
that
ap
plied
on
these
C
ADG
enes
set
.
T
his
m
anu
s
cript
dep
ic
te
d
a
novel
opti
m
a
l
featur
e
sel
ect
io
n
te
c
hn
i
qu
e
tha
t
us
es
bi
directi
on
al
po
oled
va
r
ia
nce
est
i
m
ation
(BP
VE)
for
c
oron
a
ry
artery
disea
se
pr
e
dicti
on.
Learn
i
ngs
f
rom
the
con
te
m
po
ra
ry
li
te
ratur
e
sta
ti
ng
that
existi
ng
c
la
ssifie
r
s
a
re
unsta
ble
to
ward
s
cl
assifi
cat
io
n
accu
racy
a
nd
inc
onsta
nt
t
o
la
bel
the
in
divi
du
al
record
,
he
nce
t
he
la
bel
pre
dicti
on
for
giv
e
n
r
ecord
of
the
in
div
id
ual
is
hi
ghly
false
al
a
rm
ed.
Co
ns
ide
rin
g
t
his,
a
novel
cl
assifi
er
as p
re
dicti
on
scal
e p
r
opose
d
her
e
i
n
t
his ar
ti
cl
e.
The
de
picte
d
cl
assifi
e
r
buil
t
over
t
he
swar
m
intel
li
gen
ce
te
chn
i
qu
e
cal
le
d
Cuck
oo
Searc
h
(CS
).
T
he
e
xp
e
rim
ental
st
ud
y
sta
ti
ng
t
ha
t
the
propose
d
m
od
el
BPVE&C
S
is
the
best
t
o
re
duce
dim
ension
al
it
y
of
t
he
CA
D
Gen
e
s
am
ong
t
he
m
od
el
s
f
ound
i
n
rece
nt
li
te
ratur
e
.
The
e
xperim
ental
stud
y
c
om
par
ed
the
res
ults
obta
ine
d
fro
m
pr
opos
e
d
m
od
el
with
the
r
esults
obta
ine
d
from
con
te
m
po
ra
ry
m
od
el
that
se
le
ct
s
featu
res
thr
ough
forw
a
rd
feat
ur
e
sel
ect
ion
a
nd
cl
a
ssifie
s
us
in
g
SVM
(F
FS
&S
VM)
[
28
]
.
The
propose
d
m
od
el
e
vin
ce
d
18
genes
as
optim
al
featur
e
s,
w
hich
is
best
c
ount
tha
t
com
par
ed
to
a
ny
of
th
e
c
onte
m
po
rar
y
m
od
e
l
f
ound
i
n
rece
nt
li
te
ratu
re.
T
he
la
bel
predic
ti
on
strat
egy
th
rou
gh
the
propose
d
cl
assifi
er
t
hat
bui
ld
ov
e
r
c
uc
koo
search
is
c
onsis
te
nt
in
cl
assi
ficat
ion
acc
uracy
,
evi
nced
le
ss
fa
ll
ou
t
and
m
issi
ng
r
at
e and
high se
nsi
ti
vity
an
d
sp
e
ci
fici
ty
. Th
e p
r
ocess
c
om
pletio
n
ti
m
e o
f
the
pro
po
se
d
m
od
e
l al
so
fou
nd
as
m
uch
le
ss
an
d
li
near
that
com
par
e
d
to
the
F
FS&
S
VM.
The
fu
t
ure
res
earc
h
can
exten
d
this
w
ork
to
disco
ver
the
p
ossi
bili
ti
es
of
us
ing
ot
her ANO
VA
sta
nda
rds
l
ike W
il
c
oxon
Sign
e
d
ra
nk,
E
ntr
op
y
te
st
to
r
edu
ce
the d
im
ension
a
li
ty
o
f
the
featu
re s
et
.
REFERE
NCE
S
[1]
Thom
T,
Haa
se
N,
Rosam
ond
W
,
How
ard
VJ
,
Rum
sfeld
J,
Manol
i
o
T,
Zh
eng
ZJ
,
F
le
ga
l
K,
O'
Donn
el
l
C,
Ki
tt
n
er
S,
Ll
o
y
d
-
Jones
D.
“
Hea
rt
dise
ase
a
nd
stroke
statisti
cs
--
2006
upda
te
:
a
rep
or
t
from
t
he
Am
eri
c
an
He
art
As
soci
at
ion
Stat
isti
cs
Com
mi
tt
e
e and
Strok
e S
ta
ti
s
tics Subco
m
m
it
te
e.
”
Circu
l
ati
on
.
vol.
113
,
no.
6
,
pp
.
e85
,
2
006.
[2]
Th
y
g
ese
n
K,
Al
per
t
JS
,
W
hite
HD
,
Jaff
e AS
,
Ap
ple
FS
,
Ga
lva
ni
M, K
at
us
HA
,
Newb
y
LK
,
R
avkilde
J,
Chaitm
an
B,
Cle
m
m
ense
n
PM
.
Univer
sa
l
d
e
fini
ti
on
of
m
y
o
c
ard
ial
infa
r
ct
ion
Kristia
n
Th
y
ges
en,
Jos
eph
S.
Al
per
t
and
Harve
y
D.
“
W
hit
e
on
beh
alf
of
th
e
Joint
E
SC
/ACCF
/AH
A
/WH
F
Ta
sk
Force
for
th
e
Red
ef
ini
ti
on
of
M
y
oc
ard
ial
Inf
arc
t
ion.
”
European
Heart
Journal
.
vo
l.
28
,
no.
20
,
pp.
2525
-
38
,
2007
.
[3]
Egge
rs
KM
,
L
in
d
L,
Venge
P,
L
i
ndahl
B
.
“
W
il
l
t
he
univ
ersa
l
d
e
fi
nit
ion
of
m
y
oc
ar
dia
l
infa
r
ct
ion
cr
it
eria
r
esult
in
an
over
dia
gnosis of
m
y
oc
ard
ial
infarct
ion
.”
The
Am
eric
an
Journal
o
f
Cardiolog
y
.
vo
l.
103
,
no
.
5
,
pp
.
588
-
91
,
2009
.
[4]
W
ang Z,
Luo X
,
Lu Y
,
Yang B.
“
m
iRNAs
at
the
hea
rt of
the
m
atte
r.
”
Journal
of M
ole
cu
lar
Me
dici
ne
.
vol
.
86
,
no
.
7
,
pp.
771
-
83
,
200
8
.
[5]
De
Plane
l
l
-
Sagu
er
M,
Rodi
ci
o
MC.
“
Dete
ction
m
et
hods
for
m
ic
roRNA
s
in
cl
inic
pra
ct
i
ce.
”
Cl
in
ic
al
Bi
och
emistry
.
vol.
46
,
no
.
10
,
p
p.
869
-
8
78
,
201
3.
[6]
Mela
nder
O,
Ne
wton
-
Cheh
C
,
A
lmgren
P,
Hedbl
ad
B
,
B
erg
lund
G,
Engström
G
,
Perss
on
M,
Sm
it
h
JG
,
Magnuss
o
n
M,
Christ
ensson
A,
Stru
ck
J.
“
Novel
and
conve
nt
i
onal
biomarke
rs
for
pr
edi
c
ti
on
of
inc
id
ent
c
ard
iov
asc
ula
r
eve
nts
in
the
comm
unity
.
”
Jama
.
vol.
302
,
no.
1
,
pp
.
49
-
57
,
2009.
[7]
Shah
T
,
Casas
J
P,
Cooper
JA
,
T
zoul
ak
i
I
,
Sofat
R,
McCorm
a
ck
V,
Sm
ee
th
L
,
Dea
nfi
el
d
JE,
Lo
we
GD
,
Rum
ley
A,
Fow
kes
FG
.
“
Crit
ical
appr
ai
sal
o
f
CRP
m
ea
sure
m
ent
for
the
pre
dic
ti
on
of
cor
on
ar
y
h
ea
rt
dise
ase
ev
ent
s:
n
ew
d
ata
and s
y
st
ematic r
ev
ie
w
of
31
pros
pec
t
ive
cohor
ts.
”
Inte
rnationa
l
Jo
urnal
of Ep
ide
mi
ology
.
vol.
38
,
n
o.
1
,
pp.
217
-
31
,
2009.
[8]
W
il
son
PW
,
Pen
ci
na
M,
Ja
cque
s
P,
Selhub
J,
D'
A
gostino
R,
O'
Do
nnel
l
CJ.
“
C
-
re
a
ct
iv
e
protein
and
Rec
l
assific
a
ti
on
of
Cardiovascul
ar
Risk
in
th
e
Fr
amingham
Hea
rt
Stud
y
C
li
ni
ca
l
Perspec
ti
v
e
.
”
C
i
rculat
ion:
Cardi
ovasc
ular
Quali
t
y
and
Outcome
s
.
v
ol.
1
,
no
.
2
,
pp
.
92
-
9
7
,
2008
.
[9]
Pedrotty
DM
,
Morle
y
MP
,
Ca
ppola
TP.
“
Trans
cri
ptomic
bio
m
ark
ers
of
ca
r
diova
scul
ar
dise
ase
.
”
Progress
in
Cardiovascul
ar
Diseases
.
vol
.
55
,
no
.
1
,
pp
.
64
-
9
,
2012
.
[10]
Randi
AM
,
Big
uzz
i
E,
Falciani
F,
Merli
n
i
P,
Bla
kemore
S,
Bramucc
i
E,
Luc
r
ez
io
tt
i
S,
L
enno
n
M,
Faioni
E
M,
Ardiss
ino
D,
Mannuc
ci
PM
.
“
Ide
nti
ficat
ion
of
diff
ere
nt
ia
l
l
y
expr
essed
gene
s
in
cor
o
nar
y
at
h
ero
scl
er
oti
c
p
la
ques
fro
m
p
at
i
ent
s
with
st
a
ble
or
unst
abl
e
a
ngina
b
y
cDNA
arr
a
y
ana
l
y
sis.
”
Journal
of
Thr
ombos
is
and
Haemos
tasis
.
vol.
1
,
no.
4
,
pp
.
829
-
8
35
,
2003
.
[11]
Archa
cki
SR
,
A
nghel
oiu
G,
T
ian
XL,
Ta
n
FL,
DiPaola
N,
She
n
GQ
,
Morave
c
C,
El
l
is
S,
Top
ol
EJ,
W
ang
Q.
“
Ide
nti
ficati
on
of
new
gene
s
diffe
ren
ti
a
lly
ex
pre
ss
ed
in
cor
onar
y
arter
y
di
sea
se
b
y
expr
e
ss
ion
profil
ing.
”
Phy
siologi
cal genom
ic
s
.
vol
.
15
,
no.
1
,
pp
.
65
-
74
,
2003.
[12]
El
ashoff
MR,
W
ingrove
JA
,
B
eineke
P,
Dani
el
s
SE,
Ti
ngl
e
y
W
G
,
Rosenb
erg
S,
Voros
S,
Kraus
W
E,
Ginsburg
GS
,
Sc
hwart
z
RS
,
Ellis
SG
.
“
Deve
lo
pm
ent
of
a
bloo
d
-
base
d
g
ene
ex
pre
ss
ion
a
lgori
th
m
for
assess
m
ent
of
obstruc
ti
v
e
cor
onar
y
arter
y
di
sea
se
in
non
-
d
i
abe
t
ic
p
atient
s.
”
BMC
Me
d
ic
al
G
enomic
s
.
vo
l.
4
,
no.
1
,
pp
.
26
,
20
11
.
[13]
Kitt
le
son
MM
,
S
hui
QY
,
Iri
za
rr
y
RA,
Minhas
KM
,
Edne
ss
G,
Cont
e
JV
,
Parm
igi
an
i
G,
Mill
er
LW,
Chen
Y,
Hal
l
JL,
Garc
ia
JG
.
“
Ide
nti
ficat
ion
of
a
gen
e
expr
essio
n
profile
that
d
iffe
ren
ti
a
te
s
betw
ee
n
ische
m
ic
and
no
isch
emi
c
ca
rdiom
y
opat
h
y
.
”
Circu
lat
ion
.
vo
l.
110
,
no
.
22
,
pp
.
3444
-
51
.
[14]
Kitt
le
son
MM
,
Minhas
K
M,
Iri
za
rr
y
RA,
Shui
QY
,
Edne
ss
G,
Bret
on
E,
Con
te
JV
,
Tomaselli
G,
Gar
ci
a
JG
,
Har
e
JM
.
“
Gene
expr
ession
anal
y
sis
of
ische
m
ic
and
nonische
m
ic
cardiom
y
opat
h
y
:
s
har
ed
and
dist
in
ct
gen
es
in
the
deve
lopment
of heart failure.
”
P
hysiol
ogic
a
l
Gen
omics
.
vol
.
21
,
n
o.
3
,
pp
.
299
-
30
7
,
2005
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Com
pu
t. Sci.
I
nf. Tec
hnol.
Geno
me
fe
at
ure
opti
miza
ti
on
and
c
oro
na
ry
ar
te
ry
disease
pr
e
dicti
on
us
in
g
c
uckoo
searc
h
… (
E. Neel
im
a
)
115
[15]
Min
KD
,
As
akur
a
M,
Li
ao
Y,
Na
kamaru
K,
Oka
z
aki
H,
Ta
k
aha
sh
i
T
,
e
t
a
l.
“
Ide
n
tification
of
gene
s
relate
d
to
h
ea
rt
fai
lur
e
using
glo
bal
g
ene
expr
ession
profiling
of
h
um
an
fai
l
ing
m
yoca
rdium.
”
B
ioch
and
B
iophy
Res
Comm
.
vol.
39
3,
no
.
1
,
pp
.
55
-
60
,
2010.
[16]
Suresh
R,
L
i
X,
Chiriac
A
,
Goe
l
K,
Te
r
zi
c
A,
P
ere
z
-
Te
r
zi
c
C,
Nelson
TJ.
“
Tr
a
nscript
m
e
from
ci
r
cul
a
ti
ng
cell
s
suggests
d
y
sre
g
ula
t
ed
p
at
hwa
ys
associate
d
wi
th
long
-
te
rm
r
e
cur
ren
t
ev
ent
s
foll
owing
first
-
t
ime
m
y
o
ca
rdi
al
infa
rc
ti
on.
”
Jour
nal
of
Molecul
ar
and
Ce
ll
ular
Cardiology
.
vol
.
74
,
pp
.
13
-
21
,
201
4
.
[17]
Li
ew
CC,
Ma
J
,
Ta
ng
HC, Zheng
R,
Dem
pse
y
A
A.
“
The pe
r
iphe
r
al
b
lood
tra
nscr
i
pt
m
e
d
y
namic
all
y
r
efl
e
ct
s s
y
ste
m
wide
biol
og
y
:
a
pote
ntial
d
ia
gno
stic
too
l.
”
Journ
al
of
Labor
atory
and
Cli
ni
cal
Me
dic
in
e
.
vol
.
147
,
np.
3
,
pp.
126
-
32
,
2006
.
[18]
Kaz
m
i
N,
Gaunt
TR
.
“
Diagnosis
of
cor
on
ar
y
h
eart
d
isea
ses
using
gen
e
expr
ession
profi
li
ng;
stab
le
cor
on
ar
y
ar
t
e
r
y
disea
se,
c
ard
i
ac
i
sche
m
ia
with
an
d
without m
y
oc
a
rdia
l
necros
is.
”
Pl
oS
on
e
.
vol. 1
1,
no
.
3
,
pp
.
e01
49475
,
2016
.
[19]
Yu
H,
Ni
J,
Zh
ao
J.
“
ACO
Sampl
ing:
An
an
t
c
olon
y
op
ti
m
ization
-
base
d
under
sam
pli
ng
m
et
hod
for
cl
assif
y
ing
imbala
nc
ed
DN
A m
ic
roa
rra
y
da
ta
.
”
N
euroc
omputing
.
vol. 101
,
no.
2
,
pp
.
309
–
3
18,
2013
.
[20]
Uze
r
MS
,
Yilma
z
N,
In
an
O
.
“
Fe
at
ure
sel
ection
m
et
h
od
base
d
on
ar
ti
fi
ci
a
l
bee
colon
y
a
lgori
thm
a
nd
support
vecto
r
m
ac
hine
s
for
m
e
dic
a
l
da
ta
sets
cla
ss
ifi
ca
ti
on
.
”
The
Sci
en
ti
fic
World
Journal
.
vo
l.
201
3,
no.
11,
pp.
419
187,
2013.
2013
.
[21]
Arafa
t
H,
El
aw
a
d
y
RM,
B
ara
k
at
S,
El
rashid
y
NM
.
“
Us
ing
rough
set
and
ant
co
lon
y
opti
m
iz
a
ti
on
in
f
ea
tur
e
sel
ec
t
ion.
”
Inte
rnational
Jo
urnal
of Emergi
ng
Tr
ends
&
Technol
ogy in
Com
pute
r Sc
ie
nc
e
(
IJ
ETT
CS)
.
vol.
2,
no.
1
,
2013
.
[22]
Ta
ha
AM
,
T
an
g
AY
.
“
Bat
al
g
orit
hm
for
roug
h
set
at
tr
ibut
e
r
educ
t
ion.
”
Journal
o
f
Theor
et
i
cal
and
Applie
d
Infor
mation
Tec
hnology
.
vol
.
51
,
no.
1
,
pp.
1
-
8,
2
013.
[23]
Kum
ar
PG
,
Vijay
SA
,
Deva
r
aj
D.
A
h
y
br
id
col
on
y
fuz
z
y
s
y
stem
for
ana
l
y
z
ing
di
abe
t
es
m
ic
roa
rra
y
d
at
a
.
InCom
puta
ti
ona
l
Inte
l
li
ge
nce
in
Bioi
nform
at
i
cs
a
nd
Com
puta
ti
on
al
Bio
log
y
(CIB
CB),
2013
IE
EE
Symposium
on
IEE
E
,
pp
.
104
-
1
11,
2013
.
[24]
Sathi
shkum
ar
K,
Thi
ag
ara
su
V,
R
amali
ngam
M.
A
n
eff
icien
t
ar
ti
fi
c
ia
l
b
ee
col
on
y
an
d
fuz
z
y
c
m
ea
ns
base
d
cl
ust
eri
ng
gene
expr
ession
dat
a
.
Int
ern
ation
al
Journal
of
Inn
ovat
iv
e
Researc
h
in
Com
pute
r
a
nd
Com
m
unic
at
i
on
Engi
ne
eri
ng
.
2013
Ju
l;
1(5)
.
[25]
Naka
m
ura
RY,
Pere
ira
LA,
Ro
drigue
s
D,
Cost
a
KA
,
Pap
a
JP
,
Yang
XS
.
“
Sw
arm
Inte
l
li
gen
ce
and
B
io
-
Inspire
d
Com
puta
ti
on:
9.
Bina
r
y
B
at
Algo
rit
hm
for
Fe
at
ur
e
Sel
ec
t
ion.
”
El
s
evi
er
Inc
.
Ch
apters
.
2013
.
[26]
Alsham
la
n
H,
B
adr
G,
Aloha
li
Y.
“
MRM
R
-
AB
C:
a
h
y
brid
g
en
e
select
ion
a
lgori
t
hm
for
ca
n
ce
r
c
l
assific
a
ti
on
usin
g
m
ic
roa
rra
y
gene
expr
ession
profi
l
ing.
”
Bi
o
Me
d
r
e
search
internati
onal
.
vo
l. 2015,
no.
9
,
pp
.
1
-
15
,
2015
.
[27]
Chen
KH
,
W
ang
KJ
,
Tsai ML,
W
ang
KM
,
Adria
n
AM
,
Cheng
W
C,
Yang TS
,
T
en
g
NC,
T
an
KP
,
Chang
KS
.
“
Ge
ne
sele
c
ti
on
for
cance
r
ide
n
ti
f
ic
a
tion:
a
d
ecision
tr
ee
m
odel
empow
ere
d
b
y
p
art
i
cle
sw
arm
opt
imiza
t
ion
al
gori
thm.
”
BMC
bioi
nform
at
i
cs.
vol
.
15
,
no
.
1
,
pp
.
49
,
2014
.
[28]
Shila
skar
S,
Gh
at
ol
A.
“
Fea
ture
select
ion
for
m
edi
c
al
d
ia
gnosis:
Evaluation
for
ca
rdiov
asc
u
lar
d
isea
ses.
”
E
xpe
rt
Syste
ms
wit
h
Ap
pli
cations
.
vo
l.
40
,
no
.
10
,
pp
.
41
46
-
53
,
2013
.
[29]
Li
u
H,
Li
u
W
,
Li
ao
Y
,
Cheng
L,
L
iu
Q,
R
en
X,
Shi
L,
Tu
X,
W
ang
QK
,
Guo
AY
.
“
CAD
gene
:
a
comprehe
nsi
ve
dat
ab
ase
for
cor
onar
y
ar
te
r
y
dise
ase
gen
es.
”
Nuc
l
ei
c
Acids R
ese
arch
.
vo
l.
39
,
no.
1
,
p
.
D991
-
6
,
201
1
.
[30]
Budak
H,
Ta
şab
at
SE
.
“
A
Modif
ie
d
T
-
Score
for
Feat
ure
Selecti
o
n
.
”
Appl
i
ed
S
ci
e
nce
s
and
Engi
ne
ering,
vo
l.
17,
p
p
.
845
-
845,
2016
.
[31]
Kumm
er
O,
Sav
o
y
J,
Argand
R
E
.
“
Fea
ture
sel
ect
ion
in
senti
m
ent
an
aly
sis.
”
Proc
ee
dings
of
th
e
9t
h
Conf
ere
nc
e
on
Information
R
et
r
ie
va
l
and
Applic
ati
ons,
[32]
Sahoo
PR
,
Th
eo
rems
TM.
“
Functi
onal E
qu
at
ions
.
”
W
orld
Sc
ie
nt
if
ic
,
1998.
[33]
htt
p://ww
w.sjsu.e
du/facul
t
y
/
ger
st
m
an/
Stat
Prim
er/t
-
ta
bl
e.
pdf
,
2017
)
[34]
Yang
XS
,
Deb
S.
“
Cuckoo
se
ar
ch
vi
a
Lé
v
y
fli
g
hts.
”
In
Nature
&
Bi
ologi
cal
ly
I
nspired
Computi
ng
,
NaBIC
2009
.
W
orld
Congress
on
IEEE, pp. 21
0
-
214,
2009
.
[35]
Barr
ett
T
,
W
il
h
i
te
SE
,
L
edoux
P
,
Ev
ange
l
ista
C
,
Kim
IF,
Tomashevsk
y
M
,
Marsh
al
l
KA
,
Phill
ipp
y
KH
,
Sherm
an
PM
,
Holko
M
,
Yefa
nov
A.
“
NCBI
GEO:
ar
chi
v
e
f
or
func
ti
on
al
gen
om
ic
s
data
sets
—
updat
e
.
”
Nuc
l
ei
c
ac
ids
resear
c
h
.
vol.
41
,
no
.
D1
,
pp.
D991
-
5,
201
3.
[
3
6
]
Pow
ers
DM
.
“
Eva
luation:
from
pre
ci
sion
,
r
ecal
l
and
F
-
m
ea
sure
t
o
ROC,
unfo
rcedness,
nak
edne
s
s
and
cor
re
la
t
ion
.
”
Journal
of
Mac
h
ine
Learning
Te
chnol
ogi
es
,
vol
.
2,
no
.
1
,
pp
.
37
-
63,
2011
.
Evaluation Warning : The document was created with Spire.PDF for Python.