1 computational biophysics and drug design jung-hsin lin ( 林榮信 ) division of mechanics,...

1

Computational Biophysics and Drug Design

Jung-Hsin Lin (林榮信 )

Division of Mechanics, Research Center for Applied Sciences & Institute of Biomedical Sciences,

Academia Sinica

School of Pharmacy, National Taiwan Universityhttp://rx.mc.ntu.edu.tw/~jlin/

2007/3/8 NCTU IoP Seminar


Many roles of computation in drug discovery

־ better efficiency

־ lower cost

־ better affinity to the target

־ better selectivity

־ better solubility

־ better oral availability

־ better permeability

־ better bioavailability

־ better metabolites

־ no conflict of interests

Computation can be helpful for discovering new drugs with

3

Integrated Ligand-Based & Structure-Based Virtual

Screening of Therapeutic Agents for Huntington

Disease Min-Wei Liu (劉明暐 )

An-Liang Cheng (鄭安良 )


Attenuation of GPCR Signaling


Signaling Pathways from GPCR Families


Sequence Alignment for A2A Adenosine Receptors

CLUSTALW score AA2AR_MOUSE 410 , AA2AR_RAT 410 = 95 CLUSTALW score AA2AR_HUMAN 412 2 AA2AR_MOUSE 410 = 81 CLUSTALW score AA2AR_HUMAN 412 2 AA2AR_RAT 410 = 81


Training compounds

O

N N

N

N

N

HH

NN

O

O

N N

N

N

N

HH

NN

N

HH

O

N N

N

N

N

HH

NN

OH

H

O

N N

N

N

N

HH

NN

O

NO

N

O

N

N O

O ON

N N

N

N

N

HH

NF

O

N N

N

N

N

HH

NN

OO

N N

N

N

N

HH

NN

O

N N

N

N

N

HH

NN

O

N N

N

N

N

HH

NN

O

N N

N

N

N

Cl

HH

O

N N

N

N

N

HH

NN

OH

1 2 3 4

5 6 7 8

9 10 11 12


Training compounds

O

N N

N

N

NH

H

ONO

N

O

N

N O

ONO

N

O

N

N Br

O

N N

N

N

N

HH

N

N

NO

N

O

N

N O

ONO

N

O

N

N Cl

SNO

N

O

N

NN

NN

N

N

Cl

HH

NO

N

O

N

N Cl

O

N N

N

N

N

HH

NN

SO O

O

H

NO

N

O

N

NNNO

N

O

N

N

13 14 15 16

17 18 19

20

2221 2324


Structural Alignment of General Molecules

Verapamil Carvedilol


Verapamil


Carvedilol


Pharmacophore model for A2A antagonists

Best HypoGen pharmacophore model Hypo1 aligned to compound 1


y = 0.7456x + 2.0334

R2 = 0.751

3

4

5

6

7

8

9

10

11

3 4 5 6 7 8 9 10 11

measured activity (-logKi)

estim

ated

act

ivity

(-log

Ki)

Correlation Plot


Pharmacophore model for A2A agonists

Best HypoGen pharmacophore model Hypo2 aligned to compound 33


Correlation Table


Correlation Plot

y = 0.9935x

R2 = 0.8716

3

4

5

6

7

8

9

10

3 4 5 6 7 8 9 10

measured activity (-logKi)

estim

ated

act

ivity

(-log

Ki)


Model from GPCR DB


Model from ModBase

19

A Novel Global Optimization Algorithm for Protein-Ligand

Interactions Jung-Hsin Lin (林榮信 )

Tien-Hao Chang (張天豪 )Yen-Jen Oyang (歐陽彥

正 )


Characteristics of Biological Complex Problems

• The potential energy function is extremely rugged.

• The potential energy surface is usually highly asymmetric.

• The true global minimum is often surrounded by many deceptive local minima.

• The biological complex problems are mostly in the space of high dimensionality.


The Flexible Docking Problem


Thermodynamic Process of Docking


AutoDock Scoring Function

A free energy-based empirical approach.

soltorelechbondvdw GGGGGG

22 2

,

,

,1012

,612

)(

)(

ijr

jiijjisolsol

tortortor

ji ijij

jielecelec

jihbond

ij

ij

ij

ijhbondhbond

ji ij

ij

ij

ijvdwvdw

eVSVSWG

NWG

rr

qqWG

Er

D

r

CtEWG

r

B

r

AWG

Dobs KRTG ln

J. Comput. Chem. 19: 1639-1662 (1998)


Searching is Generally a Global Optimization Problem

Usually there is no general solution. Most heuristics cannot guarantee the optimal

solution. Some of them have been classified as NP-

complete or NP-hard problem.


How to explore the phase space?(Or, how to find a needle in a haystack?)

---Importance sampling

We should only explore the important region of the phase space, not the entire phase space.

Stochastic methods usually outperform deterministic approaches in higher dimensional space.


Genetic Algorithm

1.[Start] Generate random population of n chromosomes (suitable solutions for the problem)

2.[Fitness] Evaluate the fitness f(x) of each chromosome x in the population

3.[New population] Create a new population by repeating following steps until the new population is complete a.[Selection] Select two parent chromosomes from a

population according to their fitness (the better fitness, the bigger chance to be selected)

b.[Crossover] With a crossover probability cross over the parents to form new offspring (children). If no crossover was performed, offspring is the exact copy of parents.

c.[Mutation] With a mutation probability mutate new offspring at each locus (position in chromosome).

d.[Accepting] Place new offspring in the new population 4.[Replace] Use new generated population for a further run of

the algorithm 5.[Test] If the end condition is satisfied, stop, and return the

best solution in current population 6.[Loop] Go to step 2


Chromosomes for Flexible Docking

Crossover operation

Leach, 2001.


Lamarckian Genetic Algorithm

LGA is a hybrid of the Genetic Algorithm with the adaptive local search method.

As in the GA scheme, energy is regarded as the phenotype, and the compound conformation and location are regarded as the genotype.

In the LGA scheme, phenotype is modified by the local searcher, and then the genotype is modified by the locally optimized phenotype.

In AutoDock, the so-called Solis-Wet algorithm is used (basically energy-based random move).


The Rank-based Adaptive Mutation Evolutionary Algorithm

• n individuals, denoted by s1, s2, …, sn, are generated. Each si is a vector corresponding to a point in the domain of the objective function f . In order to achieve a scale-free representation, each component of si is linearly mapped to the numerical range of [0,1].

• The individuals in each generation of population are then sorted in the ascending order based on the values of the energy function on evaluated on these individuals. Let t1, t2, … tn denote the ordered individuals and we have f(t1) ＜ f(t2) ＜ f(tn).

• n Gaussian distributions, denoted by G1, G2, … Gn, are generated before the new generation of population is created. The center of each Gaussian distribution is selected randomly and independently from t1, t2, … tn, where the probability is not uniform but instead follows a discrete diminishing distribution, n : n-1 : … : 1.

2

2

2exp

2

1)(

i

k

i

iGtx

x1

)1)((2

n

ki

Nucleic Acids Research 33: W233-W238 (2005)


The RAME Algorithm

2

2

2exp

2

1)(

i

k

i

iGtx

x1

)1)((2

n

ki


LGA versus RAME

342007/3/8 NCTU IoP Seminar http://bioinfo.mc.ntu.edu.tw/medock/, Nucleic Acids Research 33: W233-W238 (2005)


Randomized Benchmark Functions

0.1~

]1.1,9.0[~1~

]1,0[ ~128or 32, 8,

~

i

mi

i

wk

k

k

i i

im

i

i

k

i i

im

i

i

w

wf

ˆ

12

2

~

12

2

ˆ2

ˆexp

ˆ2

ˆ

~2

~exp

~2

~)(

x

xx

32/1ˆ

]1.1,9.0[ˆ

1ˆ

]1,0[ ˆ

2048or 512, 128,ˆ

i

mi

i

wk

k

m: dimensionality


Performance of LGA vs. ME for a Random Benchmark Function

Number of runs

Pro

babi

lity

of

find

ing

the

glob

al m

inim

a

0

10

20

30

40

50

60

70

80

90

100

0 2000000 4000000 6000000 8000000 10000000 12000000 14000000 16000000

ME

LGA

RLGA


Summary for the RAME Algorithm

• Our new RAME algorithm can find out the global minima for complex potential functions below dimensionality of 30 with substantial finite probability, which is suitable for most docking applications.

• The RAME algorithm avoids the “purification” effect inherent in the genetic algorithm and its derivatives, and therefore reduce the over-compression of information in the searching process.

1 computational biophysics and drug design jung-hsin lin ( 林榮信 ) division of mechanics,...

Documents