Chemistry & Biology

Innovations

In Silico Pharmacology: Computer-Aided Methods

Could Transform Drug Development

Chandra Shekhar

DOI 10.1016/j.chembiol.2008.05.001

With the rapidly growing and aging world

population comes an urgent and rising

demand for new and better drugs. Many

observers feel that the pharmaceutical

industry seems unable to satisfy this de-

mand. Although the industry’s investment

in research and development has steadily

increased, from about $16 billion in 1993

to nearly $60 billion in 2007, the number

of new molecules and biologics approved

each year has remained almost un-

changed. The past few decades have pro-

duced a dazzling array of high-throughput

life science technologies such as geno-

mics,proteomics, andmassspectrometry

that have dramatically increased the num-

ber of ‘‘druggable’’ targets. Unfortunately,

this has not yet translated into a flood of

new remedies. Indeed, as the US Govern-

ment Accountability Office points out, until

now such technologies have resulted only

in ‘‘increasing expenses without a com-

mensurate increase in the number of

drugs developed.’’

According to industry estimates, bring-

ing out a new drug typically takes up to 15

years and costs $1 billion or more. Failure

rates remain high: only 1 or 2 compounds

out of 10,000 tested make it to the market.

Many drugs may go through 75–100 clin-

ical trials and often fail at the later stages

after considerable resources have already

been spent on them. ‘‘The efficiency of

drug development is low,’’ says Carl Peck,

the founding director of the Center for

Drug Development Science at the Univer-

sity of California, San Francisco (UCSF). A

former director of the Center for Drug

Evaluation and Research at the FDA, Peck

believes that the wider use of in silico

modeling and simulation methods could

help improve this situation. Sean Ekins,

a clinical pharmacologist and editor of

a recent book on computer applications

in pharmaceutical research and develop-

ment, agrees. ‘‘The combination of chem-

istry and biology with informatics has led

to advances in silico pharmacology,’’ he

says. ‘‘In silico methods can now simulate

practically every aspect of drug discovery

and development.’’

The use of quantitative methods in

pharmacology dates back to the late

19th century, when relations between

physical and chemical properties of com-

pounds and their biological activities were

first studied. Decades later, in the 1960s,

Corwin Hansch and other scientists

began to establish so-called quantitative

structure-activity relationships (QSARs)

tying various molecular descriptors to

physical, chemical, and biological pro-

perties. This effort still continues, and

the public domain ‘‘C-QSAR’’ database,

which now contains thousands of such

models, has become a valuable resource

for drug discovery. In the 1980s and 90s,

advances in computer technology led to

a number of in silico techniques for mod-

eling ligand-target interactions. Such

drug discovery techniques have rapidly

grown in speed and power. ‘‘You can now

run all your molecules of interest simulta-

neously against different computational

models of targets and antitargets,’’ says

Ekins. ‘‘This was a pipe dream a decade

ago, but it is reality now.’’

To find new lead compounds, a conven-

tional in vitro high-throughput screening

assay will typically test a library of up to

a million compounds against the target

of interest. Besides being expensive and

cumbersome, such large-scale assays

sometimes miss valid leads and/or give

false matches. Fortunately, many virtual

screening techniques have now become

available as a complement or alternative.

A well-known example is the DOCK soft-

ware from UCSF, which uses information

about molecular structure to predict how

well a ligand will bind to a target. In

2002, researchers at Pharmacia Corpora-

tion (now part of Pfizer) used this and

other computer tools to match a library

of 235,000 compounds with protein tyro-

sine phosphatase-1B, an enzyme impli-

cated in diabetes. The search yielded

365 high-scoring molecules. Subsequent

in vitro testing showed that 127 of these

inhibited the enzyme effectively – a hit

rate of nearly 35%. Traditional high

throughput screening, in contrast, gave

a hit rate of just over 0.02%. Interestingly,

the two ‘‘hit lists’’ differed significantly: the

hits from virtual screening appeared more

‘‘drug-like’’ than the ones from real

screening (Doman et al., 2002).

Another validation of the in silico ap-

proach to drug screening came in 2003.

Research groups at two companies had

both been on the hunt for a small-mole-

cule inhibitor for the TGFb-1 receptor ki-

nase, a protein that helps form actin fibers

in cells and tissues. Of major interest in

cancer research, this protein is also impli-

cated in other fibrotic conditions such as

ocular scarring. The first group, at Eli Lilly,

found an inhibitor using conventional

‘‘wet-lab’’ assays. The other group, at

Biogen Idec, chose to go with an in silico

approach. It used software from San

Diego-based Accelrys to build a ligand

model based on a known weak inhibitor

of the kinase and then searched a com-

puter database of 200,000 compounds

for similar molecules. The search yielded

87 hits; the most promising of these

turned out to be identical to Eli Lilly’s mol-

ecule (Sawyer et al., 2003; Singh et al.,

of how you could come up with a lead

molecule without having to synthesize it,’’

says Shikha Varma-O’Brien, an associate

director at Accelrys. ‘‘This capability is of

tremendous importance to chemists.’’

In silico methods can now simulate practically every aspect of

drug discovery and development.

Chemistry & Biology 15, May 2008 ª2008 Elsevier Ltd All rights reserved 413

In parallel with techniques for drug dis-

covery, in silico methods for drug devel-

opment began to emerge in the 1970s.

‘‘Prior to the 70s, much of drug develop-

ment was empirical and sloppy,’’ recalls

Peck. Researchers began to use com-

putational methods to model the inter-

actions between drugs and biological

systems: the so-called pharmacokinetic

and pharmacodynamic processes. Since

then, computational tools relevant to

drug development have grown in scope

and sophistication. Biological systems

that can now be modeled in silico range

from a single pathway or disease to entire

cells, organs, patients, populations, or

even clinical trials. Other tools help re-

searchers navigate and mine the vast

amounts of information in the various

genetic, genomic, proteomic, and other

biology databases to build these models.

‘‘Over the years, there has been a para-

digm shift at the FDA towards greater

use of these technologies,’’ says Peck.

‘‘The agency has been encouraging the

drug industry to use them as well.’’

Several companies now specialize in

developing

innovative

computational

tools for drug development. Adding new

meaning to the term ‘‘computer mouse,’’

Foster City, California, based Entelos

offers researchers in type 1 diabetes their

favorite tool, the immune system of a non-

obese diabetic mouse—built entirely in

silico. Developed in collaboration with

the American Diabetes Association, the

virtual mouse immune system is far easier

to manipulate than its flesh-and-blood

counterpart. As a versatile and inexhaust-

ible resource, this in silico animal model

could open new avenues for diabetes

research, says Richard Kahn, the diabetes

association’s chief scientific and medical

officer. ‘‘There are many questions in dia-

betes that we may never resolve using

clinical trials,’’ says Kahn. ‘‘Maybe mathe-

matical models could provide effective

and accurate answers.’’

Entelos uses a top-down approach to

construct disease platforms and virtual

patient populations, drawing from clinical

trial results, pathway databases, and

other results from the literature. ‘‘We build

mechanistic models in a compartmental

way and then add granularity to those

compartments and integrate them ac-

cording to the question being asked’’

says Mikhail Gishizky, the company’s

chief scientific officer. ‘‘We don’t need to

take the modeling down to the gene level

in every area.’’ Richard Ho, a principal at

La Jolla, California, based Rosa, supports

this principle. ‘‘Stuff that happens at the

cellular level often doesn’t make it to clin-

ical significance,’’ he says. In his previous

tenure as a diabetes researcher at John-

son & Johnson, Ho used software from

Entelos to show that an experimental dia-

betes drug, which another company had

successfully tested on animal models,

would fail in humans. Meanwhile, the other

company went ahead with clinical trials

for this drug. Two years later the results

of the trial were in: the drug was ineffec-

tive. ‘‘The results looked exactly like what

we’dpredicted withourmodels,’’saysHo.

‘‘That’s lots of time, effort, and expense

we saved with four months of computer

modeling.’’

Taking a different tack, Cambridge,

Massachusetts, based Gene Network

Sciences uses massively parallel com-

puters to ‘‘reverse-engineer’’ disease

models directly from gene chip and other

high-throughput data. This data-driven

approach is particularly effective at cap-

turing the effect of genetic and genomic

variations, says the company’s executive

vice president and co-founder Iya Khalil.

She mentions results of unpublished

study where researchers used the com-

pany’s software to reverse-engineer and

simulate the action of a cancer drug;

among the genes that the simulation de-

tected as being important for the drug’s

efficacy were several that were not previ-

ously known to affect the drug’s pathway

(Pitluk and Khalil, 2007). ‘‘Using this sys-

tem, you can test millions of hypotheses

and then go back and see where these

predictions fall in the context of the litera-

ture,’’ says Khalil. ‘‘Often you end up

discovering new biology relevant for the

efficacy of the drug.’’

The list of companies developing in sil-

ico tools for pharmacology has steadily

grown during the past two decades. This

proliferation of tools and vendors has

brought its own challenges, such as in-

compatible data formats, which some

newer platforms are trying to address.

Meanwhile, at the big pharmaceutical

firms, the intended clients for these prod-

ucts, interest in in silico methods is grow-

ing. Most of these companies are now

either developing such methods in house

or working with partners that provide

them. The FDA, too, is collaborating with

many in silico tool developers including

Entelos, UK-based SymCyp, and Moun-

tain View, California, based Pharsight.

‘‘In silico methods are having an increas-

ingly important role in different areas of

pharmaceutical research and develop-

ment,’’ says Ekins.

Will computational methods ever be

able to completely replace in vitro and in

vivo testing? ‘‘The answer here can only

be a clear and resounding ‘no’, at least in

thenearfuture,’’saysEkins, alongwithco-

authors Jordi Mestres and Bernard Testa

in a pair of recent review articles (Ekins

biological systems have a highly nonlin-

ear, even chaotic nature, whereby even

tiny changes in initial conditions could

make them behave in a dramatically differ-

ent manner. ‘‘No computer program will

ever be able to fully model their complex-

ity,’’ they argue. Peck is more optimistic.

‘‘There may come a time, perhaps 50 or

100 years in the future, when drugs will

be discovered in silico, tested in silico, op-

timized in silico, and made available to

patientsforclinicalusewithlittleornocon-

firmatory testing,’’ he says. ‘‘This is a bold

and even somewhat frightening vision, but

someday we’ll know enough about biol-

ogy to go a long way towards this goal.’’

REFERENCES

Doman, T.N., McGovern, S.L., Witherbee, B.J.,

Kasten, T.P., Kurumbail, R., Stallings, W.C., Con-

nolly, D.T., and Shoichet, B.K. (2002). J. Med.

Chem. 45, 2213–2221.

Ekins, S., Mestres, J., and Testa, B. (2007a). Br. J.

Pharmacol. 152, 9–20.

Ekins, S., Mestres, J., and Testa, B. (2007b). Br. J.

Pharmacol. 152, 21–37.

Pitluk, Z., and Khalil, I. (2007). Drug Discov. Today

12, 924–930.

Sawyer, J.S., Anderson, B.D., Beight, D.W., Camp-

bell, R.M., Jones, M.L., Herron, D.K., Lampe, J.W.,

McCowan, J.R., McMillen, W.T., Mort, N., et al.

(2003). J. Med. Chem. 46, 3953–3956.

Singh, J., Chuaqui, C.E., Boriack-Sjodin, P.A., Lee,

W.C., Pontz, T., Corbley, M.J., Cheung, H.K., Ardu-

ini, R.M., Mead, J.N., Newman, M.N., et al. (2003).

Bioorg. Med. Chem. Lett. 13, 4355–4359.

Singh, J., Ling, L.E., Sawyer, J.S., Lee, W.C.,

Zhang, F., and Yingling, J.M. (2004). Curr. Opin.

Drug Discov. Devel. 7, 437–445.

Chandra Shekhar is a science writer based in

Princeton, NJ.

Chemistry & Biology

Innovations

414 Chemistry & Biology 15, May 2008 ª2008 Elsevier Ltd All rights reserved