Transcript On the nature of cavities on protein surfaces: Application to the

On the nature of cav
i ti es on prote
i n
surfaces: Appli cati on to the
Identi fi cati on of drug-b
i nd
i ng s
i tes
Murad Nayal, Barry Honig
Columbia University, NY
Proteins: Structure, Function and
Bioinformatics, Accepted 15 Nov. 05
Ankur Dhanik
Abstract
• Identification of drug-binding sites useful for
virtual screening and drug design.
• Small ligands are known to bind proteins at
surface cavities.
• Two tasks: identification of cavities and
prediction of their drugabbilities (whether the
cavity is suitable for drug binding).
• The method presented in this paper encoded in
program called SCREEN (Surface Cavity
REcognition and EvaluatioN).
Abstract
• SCREEN works by first constructing two
molecular surfaces using GRASP: a conventional
molecular surface (MS) using a 1.4 A radius and a
second low resolution envelope using a large
probe sphere, which serves as ‘sea-level’. Depth
of each vertex of MS is computed and compared
with threshold.
• For each surface cavity, 408 attributes are
computed (physiochemical, structural, and
geometric).
• Random Forests based classifier is used.
• Training data set is derived from a collection of
100 nonredundant protein ligand complexes.
Results
• SCREEN predicts drug binding cavities with a balanced
error rate of 7.2% and coverage of 88.9%, while a CASTp
( a popular protein cavity detection program) based
druggability predictor (using cavity size criteria alone)
predicts with a balanced error rate of 15.7% and
coverage of 71.7%.
• SCREEN predicts drug-binding cavities missed by cavity
size criteria (three examples).
• Out 18 attributes out of 408 used, were found to be
significant predictors of drug binding cavities.
• It follows from the above that drug binding cavities are
large, deep, have an intricate curvature profile, are rigid,
and have a relatively small number of prolines, as well as
amino acids with small but negative octanol-to-water
transfer free energies (Asn, Gln, Glu).
Results
Protein-tyrosine phosphatase
1B, PTP1B (PDB code: 1l8g).
The largest surface cavity
(colored green: area, 184 Å2;
volume, 400 Å3; residues
Gln78, Arg79, Ser80, and
Pro210) is about 20 Å from the
ligand-binding site. The drug
binds at the second largest
cavity, colored red, as
predicted (area, 170 Å2;
volume, 259 Å3; residues
Gln262, Ala217, Ile219, Val49,
and Asp181).
Results
Human carbonic anhydrase II
(CA II). The largest cavity
(area, 281 Å2; volume, 679 Å3;
residues Phe213, Tyr7, Gly8,
Asp243, and Lys170), shown in
green, is rather shallow and is
predicted not to bind a drug.
Instead, the second largest
cavity (area, 194 Å2; volume,
281 Å3; residues Leu198,
Thr200, His94, Val121, and
His64) is the one predicted
correctly to bind the drug.
Results
Human factor Xa complexed with
inhibitor RPR128515 (PDB code:
1ezq). Four cavities ranked 1, 2, 3,
and 9, shown here in red, were
predicted to be potential drug-binding
cavities. The ligand actually binds at
two cavities, 3 (the S1 pocket: area,
274 Å2; volume, 384 Å3; residues
Gln192, Trp215, Ser195, Cys191,
Gly216, and Asp) and 9 (the S4
pocket: area, 69 Å2; volume, 155 Å3;
residues Trp215, Phe174, Thr98,
Tyr99, and Ile175).
R
D
S
u
r
C
a
v
f
a
i
c
t
e
y
c
r
a
a
v
n
i
t
y
p
r
o
p
e
r
t
y
C
a
t
e
g
o
r
y
r
u
g
c
-
a
b
v
i
e
i
n
t
i
d
i
e
n
s
g
u
N
s
b
i
n
o
d
i
n
n
l
d
g
r
u
c
a
g
v
u
m
b
e
r
o
f
r
N
u
m
b
e
r
o
f
a
S
m
e
s
t
i
o
d
u
m
e
s
s
S
i
z
e
1
.
S
i
z
e
2
2
S
i
z
e
8
5
9
±
2
.
.
8
±
1
4
.
0
±
6
2
8
l
-
i
t
i
e
s
7
8
.
8
8
±
5
.
4
.
3
7
.
3
1
±
5
.
4
.
4
0
1
8
.
±
7
4
l
s
k
N
a
t
e
s
t
m
o
m
e
n
t
o
f
i
n
e
r
t
i
a
S
i
z
e
/
s
h
a
p
e
1
.
×
7
1
2
.
5
1
.
×
2
1
±
0
4
1
1
.
2
3
±
0
×
2
8
.
3
×
3
0
1
0
3
D
e
M
p
a
A
v
t
x
e
i
r
h
s
m
u
a
g
t
a
n
m
e
d
a
d
d
e
e
d
e
r
p
p
t
d
t
d
e
v
i
a
t
i
o
n
h
h
S
i
z
e
/
s
h
a
p
e
S
i
z
e
/
s
h
a
p
e
S
i
z
e
/
s
h
a
p
e
2
.
1
0
5
±
3
.
.
3
1
.
±
5
±
1
4
1
(
.
.
9
Å
0
)
(
(
Å
Å
)
)
0
.
7
5
±
0
.
4
5
4
.
7
5
±
1
.
6
7
2
±
0
.
7
3
.
Shape
17.0 ± 11.7
3.9 ± 5.3
Shape
0.02 ± 0.013
0.003 ± 0.001
Size/shape
1.6 × 104 ± 8.4
× 104
2.8 × 103 ± 1.6 ×
104
Average side-chains residual
entropy
Rigidity
-0.41 ± 0.18
(kcal)
-0.55 ± 0.25
Average curvature
Shape
-49.0 ± 8.3
-57.0 ± 13.1
Maximum curvedness
Shape
6.4 ± 2.9
4.0 ± 4.9
Maximum mean curvature
Shape
5.3 ± 2.6
3.5 ± 4.2
N
i
o
n
r
e
m
r
t
a
i
l
i
z
s
m
a
l
l
e
s
t
m
o
m
e
n
t
o
f
a
Proportion of cavity at depth
between [6.5, 6.75)
Largest moment of inertia
Comments
• The prediction of drug-binding cavities was done
without considering the nature of the drug.
• Physicochemical cavity properties were not
found useful.
• Perhaps they can play an important role when
surface cavities that recognize a particular
ligand are characterized.
• Energy-based approach offers a promising
alternative to geometry-based methods