In Silico Drug Design Feeds Drug Development - Biomedical ...

20 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org

DOCK THIS:Drug DesignFeeds Drug DevelopmentBY KRISTIN COBB, PHDOnce upon a time, not long ago, HIV/AIDS was a scourge, killing anyonewho contracted the deadly virus. Now, many people are living withthe disease, which they control with drugs initially developed in the 1980sand early 1990s using an approach called computer-aided drug design—the use of computer models to find, build, or optimize drug leads.Armed with information about the 3-D structure of HIV protease, anenzyme essential to the HIV reproductive cycle, computationalresearchers designed molecules in silico toprecisely fit the shape of theenzyme’s active site—asthough fitting a key to alock. The resultingdrugs, potent inhibitorsof HIV protease and theHIV life cycle, werebrought to market in recordtime and revolutionized the treatmentof HIV/AIDS.Around the same time, another anti-viral—Relenza, which treatsinfluenza and was a forerunner to Tamiflu—was also designed using thesemethods. These HIV and flu drugs are among the best known success storiesof computer-aided drug design (see page 23 for both stories).Since those early successes, computer modeling has become an integralpart of drug discovery. “Almost everything that has recently moved forwardfrom big pharmaceutical companies to market has involved somesort of collaboration with computational chemistry. It’s like asking, werethere chemists involved? Of course there were. It is part of the process,”says Tara Mirzadegan, PhD, head of the computer-aided drug designgroup at Johnson & Johnson.www.biomedicalcomputationreview.orgSummer 2007 BIOMEDICAL COMPUTATION REVIEW 21

DOCK THIS:Drug Design Feeds Drug Development“Almost everything that has recently moved forward from bigpharmaceutical companies to market has involved some sortof collaboration with computational chemistry. It’s like asking,were there chemists involved? Of course there were. It is partof the process,” says Tara Mirzadegan.Quite often, computers play a rolewithout making the big splash they didwith Relenza and the proteaseinhibitors. That’s probably because nodrug is created solely in silico; the computeris just one of many tools in thisprocess. But as algorithms evolve, computingpower explodes, and scientistssolve a greater number of 3-D proteinstructures, computer-aided design hasthe potential to dramatically cut thecost and time of drug discovery. How?By narrowing down the field of compoundsthat might help treat a particulardisease; by assembling novel drugmolecules to disrupt specific diseasepathways; and by providing new attackroutes against traditionally difficultdrug targets. Computers are alsoincreasingly playing a role in optimizingdrug leads for bioavailability and safety.Despite the over-hype of computersas the saviors of drug developmentcompanies, many still expect thisprocess to bear important fruit.Computer-aided drug design played acritical role in the design of severaldrugs that are now in late preclinicalor early clinical development. Onlytime will tell which of these, if any, willemerge as drug success stories.Docked Drug. This 3-dimensional computer graphic shows a candidate drug (a JAK2inhibitor) docked in the active site of its target protein (JAK2). JAK2 protein is implicatedin various myeloproliferative disorders (diseases that produce excess bone marrow cells,such as chronic myelogenous leukemia, or CML) estimated to affect 80,000-100,000 peoplein the U.S.. Courtesy of SGX Pharmaceuticals, Inc.VIRTUAL SCREENINGHow it works: In the ideal situation,the 3-D structure of the target molecule(usually an enzyme or receptor) isknown, allowing scientists to directlyvisualize drug-target interactions in silico.Structure-based methods haveevolved in two directions since Relenzaand the HIV proteases—virtual screeningand fragment-based design.In virtual screening, the 3-D structureof a target is screened againstlibraries of potentially active small molecules.The computer “docks” eachcompound, or ligand, into the target’sactive site and scores its geometric andelectrostatic fit.Considerable progress has beenmade in docking programs in the lasttwo decades, but scientists agree thatthe problem is complex and that theyhave yet to find a perfect solution. Tostart with, the ligand and protein targetare often pictured as a rigid lock andkey—but in fact they are dynamic, movingobjects that continually changeshape and adjust their shapes inresponse to each other.“Imagine taking a fluffy ball and tryingto mold it to optimally fit some kind of abinding site. There are just way too manyconfigurations,” says Dimitris K.Agrafiotis, PhD, vice president ofContinues on page 2422 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org

EARLY EXAMPLES: ANTI-VIRAL DRUGSRelenza and the HIV protease inhibitors stand out asthe two classic examples of computer-aided drug design.Relenza was developed through a collaboration ofAustralian scientists, including Jose N. Varghese, PhD,head of structural biology at CSIRO Molecular and HealthTechnologies. In 1983, Varghese and his colleagues usedX-ray crystallography to solve the 3-D structure of theenzyme neuraminidase, one of two potential protein targetson the surface of flu. Neuraminidase plays a criticalrole in the flu life cycle: after the virus replicates within ahost cell, neuraminidase releases the newly formed viralprogeny by cleaving a bond between the viral surface proteinhemagglutinin and a sugar on the host cell surface,sialic acid.A series of structural experiments revealed importantinsights. The active site of the enzyme was highlyconserved in all strains of flu—bothhuman and animal; the virus routinelyescaped antibody recognition bymutating around the periphery ofthe active site but never changingthe active site itself.“Because it was so highlyconserved, it seemed clear to usthat it must have a very importantfunction,” Varghese says. “So, clearlyif one made a molecule that went inthere and blocked that site, it would be prettyeffective.”A synthetic analog of sialic acid was known to inhibitneuraminidase, but without sufficient potency. Using thecrystal structure of neuraminidase bound with this analog,the researchers set out to design a better inhibitor insilico. Computer predictions revealed that a particularguanidinium-for-oxygen substitution would give tightbinding. Synthesis of this compound—Relenza—turnedout to be tricky, but eventually succeeded.“It bound in nanomolar binding, so it was very tight,and it certainly blocked the virus replication right downto its tracks,” Varghese says.Relenza was licensed to GlaxoSmithKline Inc. in 1990and approved by the FDA in 1999. Following theirlead—and capitilizing on a patent oversight, accordingto Varghese—Gilead Sciences developed the betterknownneuraminidase inhibitor, Tamiflu (marketed byRoche). Both drugs may be important in the fightagainst bird flu, Varghese says.Development of the HIV protease inhibitors laggedbehind that of the neuraminidase inhibitors by severalyears, but the former won FDA approval sooner (in themid-1990s) because of the pressing medical need.Dale Kempf, PhD, who is now a distinguishedresearch fellow in Global Pharmaceutical Research andDevelopment at Abbott, was involved in Abbott’s developmentof ritonavir (brand name Norvir), which startedin late 1987.“It’s one of the first examples of the application ofgenomics for drug design,” he says. When the HIVgenome was sequenced and published in the mid-1980s, several groups recognized characteristicsequences suggestive of a protease enzyme.Interestingly, the gene encoded only half a protein,which led Kempf and others to realize that the proteasemust be composed of a dimer—two identical halves thatcome together to form one active site. This provideda key structural insight even beforeX-ray crystal structures of the proteasewere available: the active site hadto have a particular type of symmetry,known as C2 or two-foldsymmetry (rotation 180 degreesaround a central axis yields theidentical structure).Kempf’s group used that insightto create a computer model of theprotease active site and to design possibleinhibitors in silico by starting with a known substrate,chopping off half of the substrate, and rotatingthe remaining half by 180 degrees.“And when we went into the lab and made thosecompounds, they turned out to be very potentinhibitors,” Kempf says.Using a combination of the X-ray crystal structures ofHIV protease (which had since become available) andcomputer graphics, they modified these compounds insilico to visualize how certain substitutions wouldimprove characteristics like bioavailability. The first compoundwith sufficient oral bioavailability, ritonavir, wassynthesized in 1991.In 1996, the FDA approved ritonavir in record time(72 days). The total development time—about eightyears—was roughly half that of a typical drug, due bothto the structure-based approach and to the FDA’s acceleratedreview. Several other HIV proteases emergedaround the same time, including saquinavir (Roche) andnelfinavir (developed by Agouron, now a subsidiary ofPfizer). These drugs helped to revolutionize the treatmentof HIV.www.biomedicalcomputationreview.orgSummer 2007 BIOMEDICAL COMPUTATION REVIEW 23

DOCK THIS:Drug Design Feeds Drug DevelopmentCancer Interrupted. This three-dimensional computer graphic shows a drug candidate (MET tyrosine kinase inhibitor) bound to its target protein.MET receptor tyrosine kinase controls cell growth, division, and motility and is implicated in a range of cancers, including renal cell carcinoma,gastric cancer, lung cancer, glioblastoma and multiple myeloma. Courtesy of SGX Pharmaceuticals, Inc.Continued from page 22informatics at Johnson & JohnsonPharmaceutical Research & Development.“Small molecules—unless they’revery small—tend to be very flexible. Theyflop around a lot. They can assume a multitudeof conformations in 3-D.” If a moleculehas five rotatable bonds, then eachbond can rotate at many different angles,creating a lot of freedom to take onunique conformations.Most docking programs nowaccount for the flexibility of the ligandby sampling its many conformationsand docking each one, but adequatelyaccounting for the flexibility of the targetprotein is a much more challengingproblem. Adding protein flexibilityexponentially increases computingdemands.“The state of the art today is comingup with sensible simplifications thatmake the problem computationallytractable but still meaningful,”Agrafiotis says.Besides the flexibility of the protein,many docking programs do not adequatelyaccount for the influence ofwater—which surrounds all molecules inliving systems. “The mathematical modelsfor defining water and how it shapesitself around the receptor and the drugmolecule are still pretty unclear,” saysKent Stewart, PhD, a research fellowin structural biology at Abbott.In addition, the algorithms estimatebinding energies using classicalNewtonian physics, rather than quantumphysics—which also reduces accuracy.“You can calculate the binding energiesfrom some sort of Newtonian pointof view, treating atoms as sort of ballsattached to springs. Or you can treat itfrom a quantum mechanical point ofview. Now the quantum mechanical calculations,as you can imagine, are horrendous,”says Jose N. Varghese, PhD,head of structural biology at CSIROMolecular and Health Technologies.“At this stage, it is a computational challenge.”Methods of scoring how well a smallmolecule fits a protein’s active site alsomust trade off between speed and accuracy.“The scoring function that we usehas many shortcuts and approximations,”says Mirzadegan. Her group willvirtually dock the company’s one millionproprietary compounds (which ithas purchased or developed over theyears) against a given target, and pickthe highest ranked 10,000 for biologicaltesting. “We cannot afford docking onecompound per day. That would be one24 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org

“The state of the art today is coming up with sensible simplificationsthat make the problem computationally tractable butstill meaningful,” says Dimitris K. Agrafiotis.million days. So we have to do it in amatter of seconds or sub-seconds.”But increased computing power canhelp boost the speed of virtual screeningwithout compromising accuracy. In2000, for instance, Arthur J. Olson,PhD, professor of molecular biologyand director of the Molecular GraphicsLaboratory at The Scripps ResearchInstitute, started the FightAids@Homeproject, which uses internet-based gridcomputing—as was popularized by theSETI@Home project—to do virtualscreening for new anti-HIV drugs.“If most people who have computersuse only about five percent of theCPU cycles—and the rest of the cyclesare just idle—how much wasted or availablecomputing is there?” Olson asks.“It turns out to be an amazing number.”His grid computing project makes useof that idle computer time and helpsevaluate drugs for dealing with HIVproteins’ habit of rapidly mutating toescape drug pressures. Fortunately, the3-D structures have been solved formany of the mutant HIV proteins.With the help of about 500,000 volunteercomputers, Olson used AutoDock(a popular docking program that wasdeveloped in his lab) to screen 2000small molecules against several hundreddifferent HIV protease mutants. Theprogram took six months to run; heestimates that on the Scripps supercomputer, with 300 processors running,it would have taken 50 years.Besides identifying several drugleads, which are now in testing, Olsonrecognizes an even more important payoff:“When you do such massive dockings,you actually are collecting morethan just an answer; you’re collecting alot of statistics.” Such data could, forexample, be used to identify a subset ofmutants that represent a spanning set—Anti-Cancer Key. An anti-cancer drug compound—nutlin—bound to the cancer-causing proteinMDM2. Courtesy of RMC Biosciences, Inc.one that captures all unique interactionswith the ligands screened. “Doingdocking on only this subset of mutantswould free up computer time for screeninglarger libraries, using more dynamicrepresentations of the protein targets,or using more accurate scoringfunctions,” he says.The Folding@Home project atStanford also uses grid computingfor drug design. Led by Vijay S.Pande, PhD, associate professor ofchemistry and of structural biology,Folding@Home focuses on simulatingprotein folding and misfolding, but “asour work matures, we have been lookinginto the next steps involved in computationaldrug design,” Pande says.Using distributed computing, his grouphas devised new, more accurate algorithmsfor docking and for calculatingligand-protein binding energies. Thesealgorithms are being used in the designof several new drugs, including newinhibitors of the cytokine-cytokinereceptor interaction (involved in cancer);novel chaperone inhibitors (alsoinvolved in cancer); and novel antibioticsthat target the bacterial ribosome.“Distributed computing is a keywww.biomedicalcomputationreview.orgSummer 2007 BIOMEDICAL COMPUTATION REVIEW 25

DOCK THIS:Drug Design Feeds Drug DevelopmentFragment-based design. Drug companies, such as SGX pharmaceuticals, screen hundreds of fragments in their fragment libraries and identifyhits that serve as the building blocks for novel drug candidates. Knowledge of the binding mode of each fragment to its target is combinedwith advanced computational tools to produce “engineered” drug leads. For example, in this series, a hit is first identified throughcrystallographic screening (yellow); then chemical groups (red and pink) are added to the bound fragment to increase its binding affinity.Courtesy of SGX Pharmaceuticals, Inc.Distributedcomputing is key todeveloping better,more accuratealgorithms forcomputer-aideddrug design, saysVijay Pande. “Itallows us to docalculationsotherwiseimpossible.”aspect to this, as it allows us to do calculationsotherwise impossible,”Pande says.FRAGMENT-BASED DESIGNFragment-based methods take a“Lego” approach to drug design. In alab, scientists create chemical librariesof small compounds, or fragments—perhapsone-third the size of a typicaldrug—that are easily linked together.They then screen the libraries for bindingactivity experimentally, using highthroughputX-ray crystallography (orNMR or mass spectrometry); when afragment binds to the target, the crystallographyprovides an exact 3-D pictureof the bound fragment in the activesite. Next, with the help of computermodeling, fragments are turned intopotent drug leads by adding new chemicalgroups to the initial core fragmentor by stitching together several fragmentsthat bind to different points inthe active site.“I think this approach is showingquite good promise,” Varghese says.“In fact, with the advent of these modernsynchrotrons, scientists can do thisfairly quickly—and a lot of pharmaceuticalcompanies are moving in thisdirection.”The approach offers a combinatorialadvantage: “Instead of having a databaseof say four million compoundsthat a really large company would have,you take compounds that are say onethirdof the size, and explore them combinatorically.If you explored ten fragmentsin three different positions,you’d actually explore 1000 combinations.So with a database of somethinglike 400 compounds, you can explore achemical space that is in the several millions,”says Sir Tom Blundell, FRS,FMedSci, professor and chair of biochemistryat the University ofCambridge. In 1999, Blundell cofoundedAstex Therapeutics to do fragment-basedmethods; the company isnow testing a kinase inhibitor—a type ofcancer drug—in clinical trials.“The experiment is really one ofusing crystallography to do yourscreening. So you’ve pushed the crystallographytechnology to the pointwhere you can do it so rapidly that itbecomes effective to use as a screeningtool,” says Siegfried Reich, PhD, vicepresident of drug discovery at SGXPharmaceuticals, another companythat uses fragment-based methods.(Reich previously helped develop theHIV protease inhibitor nelfinavir atAgouron.) When it was founded in1999, SGX was named StructuralGenomix and its aim was to use highthroughput X-ray crystallography tosolve a record number of protein structures.But this was not sustainable as abusiness model. So, in 2000, the com-26 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org

“When you’re talking about toxicity, it’s much easier to give acompound to a rat than it is to dock against all possible proteinsthat are in the rat, even today,” says Art Olson. “But someday, youmight be able to do that. We’re certainly creeping up on that.”pany changed its name to SGXPharmaceuticals and put its crystallographypower to use in drug discovery.One of their lead candidates is a newinhibitor of BCR-ABL, a perpetuallyactive kinase enzyme involved in chronicmyelogenous leukemia, or CML. TheBCR-ABL inhibitor Gleevec has hadenormous success in treating CMLpatients, but 20 percent are resistant toGleevec. So scientists at SGX cloned,expressed, purified, and crystallized theGleevec-resistant protein. Then theyscreened their fragment library againstthe wild type and mutant versions ofBCR-ABL to find compounds activeagainst both. The fragment hit that eventuallyled to their lead candidate startedwith a low binding affinity of just 10micromolars (i.e., a fairly high concentrationof compound was required tobind at least half the protein).This is where the medicinal chemistsand structural biologists sit down withthe computational chemists, Reich says.Computational chemists virtually buildnew compounds by adding chemicalgroups to the starting fragment. Forexample, they might try linking all thedifferent simple alkyl amines to one ofthe fragment’s “chemical handles” (siteson the fragment that easily bind toother chemical groups), Reich explains.The computer calculates the bindingaffinity for each iteration, until it findsone with tight binding. Specialized versionsof docking programs are used tocalculate the binding affinities. Butbecause you already know exactly howthe fragment binds, you start with moreinformation than in virtual screening.By elaborating their initial lead in thisway, SGX got their first hit down tonanomolar potency—i.e. very little of thecompound was required in order to bindthe protein—in about three months.“That gives you a flavor for how fast thiscan go,” Reich says.Tricky Target. This computer model of abacterial cell membrane helped scientistsat Polymedix design new antibiotics thatmimic the action of the defensin proteins(natural proteins in the body that killbacteria by puncturing their membranes).Courtesy of Polymedix.TRICKY TARGETSDocking algorithms and fragmentbasedmethods work well on solubleenzymes that are easily crystallized andcontain well-defined pockets where ligandscan bind—but many diseasesinstead involve membrane-bound receptorsor protein-protein interactions.Membrane-bound receptors transmitsignals from outside to inside the cell.Because the proteins are embedded inthe membrane, they cannot easily becrystallized and it is difficult to solvetheir structures. For example, 25 percentof the top 100 drugs on the market todaytarget G-protein coupled receptors—including the dopamine and serotoninreceptors in the brain—but the structureof only one mammalian G-protein coupledreceptor is known.When structural information isunavailable, computational chemists useligand-based methods to hunt for newdrug leads. They superimpose a set of ligandswith known activity against the targetand compare their structural andchemical features. A common pattern,called a pharmacophore, emerges—keyfunctional groups (such as hydrogenbond donors, electrostatic charges, andhydrophobic patches) must be in certainpositions. This fingerprint is then usedto virtually screen libraries for novelcompounds with similar patterns.Ligand-based methods pre-date the structure-basedmethods and have helpeddevelop many drugs, including drugs totreat high blood pressure, pain, anddepression.Protein-protein interactions occur viasurfaces that are often featureless andshallow, and binding affinities can bequite large—so it’s hard for small moleculesto disrupt these interactions, saysArthur Olson of Scripps ResearchInstitute. You have to find or designdrugs that can bind to multiplefootholds, or hot spots, on the proteinsurface, which is challenging, he says. “Ithink that this is an area that is really stillin its infancy.”But some progress is being made.Kent Stewart of Abbott Labs hopes tocontrol BCL-2, a protein that is overexpressedin certain cancers. It blocksapoptosis (programmed cell death) andthus keeps cancer cells alive. Comparedto HIV, Stewart says, which has an actualcave you can dock a molecule into, onwww.biomedicalcomputationreview.orgSummer 2007 BIOMEDICAL COMPUTATION REVIEW 27

DOCK THIS:Drug Design Feeds Drug DevelopmentCancer Interference. The oncogenic protein BCL-2 helps keep cancer cells alive via a protein-protein interaction. This Bcl-2 inhibitor—developedat Abbott using a fragment-based approach—binds to the BCL-2 protein surface and disrupts the protein-protein interaction. The compoundis in late preclinical development. Courtesy of Abbott.BCL-2, “there’s no such thing as a cave;it’s a very flat and open surface, so it’shard to get molecules that actuallystick,” So, using a fragment-basedapproach, scientists at Abbott linkedtogether two fragments that bind to theBCL-2 protein surface, resulting in apotent compound that can disrupt theprotein-protein interaction. The compoundis now in late preclinical development.Some companies have made thesedifficult targets their niche area. Forexample, Polymedix’s mission is todevelop drugs against membrane-boundtargets, protein-protein interactions,and membrane-protein interactions,using a suite of computational toolsspecifically developed for these aims (byprofessors William DeGrado, PhD,and Michael Klein, PhD of theUniversity of Pennsylvania).Polymedix is working on a new lineof antibiotics that mimic the action ofdefensins—natural proteins found inthe body that kill bacteria.“They work similarly to a needle or acorkscrew going into a balloon. Theydirectly attack and perforate the bacterialcell membrane,” says NicholasLandekic, MBA, President, CEO, andco-founder of Polymedix. Because theydo not target bacterial proteins—whichcan easily evolve to escape drug pressures—defensin-likedrugs should notengender bacterial resistance, he says.Scientists at Polymedix built a computationalmodel of a defensin proteininserted into a bacterial cell membrane(a peptide-membrane interaction).Then they virtually transformed thedefensin protein into a drug-sized compound.By swapping amino acid groupsfor chemically analogous small moleculegroups, they shrunk the proteinwhile preserving its chemical interactions(electrostatics, lipophilicity, etc.)within the membrane.The result: drug leads one-tenth thesize of the defensins, but about 100-foldmore potent and 1000-fold more selective.“So we’ve been able to improve onnature,” Landekic says. The compoundsare now being tested in animal studies.“We’ve spent less than 14 million dollarsto date since starting Polymedix, soin terms of an efficiency and efficacyrate, I think that’s pretty good,” he adds.MAKING CHEMICALSINTO DRUGSComputer-aided methods can identifydrug leads with potent activity againsta target, but these compounds are farfrom being drugs. Drugs must also bebioavailable and safe. Safety problemsderail many drugs late in development,so identifying potential safety snagsearly on could save considerable timeand money.28 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org

HIV Protease Inhibitor. The second-generation HIV protease inhibitor, Kaletra, was developed at Abbott. Here Kaletra is shown bound to theactive site of HIV protease. Courtesy of Abbott.“How well can we evaluate bioavailabilityand toxicity in silico? It’spretty blunt and not a very popularanswer: we don’t do very well,”Stewart says. “The biological mechanismsunderlying bioavailabilityand toxicity are complex. So the mathematicalmodels in those areas are stillin their infancy,”Olson agrees: We are a long wayfrom being able to simulate a drug’seffect on the entire human body.“When you’re talking about toxicity, it’smuch easier to give a compound to a ratthan it is to dock against all possibleproteins that are in the rat, even today,”he says. “But someday, you might beable to do that. We’re certainly creepingup on that.”Computers do play a role today,however. Drugs must meet propertiesthat fall under the ADME acronym: beAbsorbed by the body, Distributed tothe target tissues, and not Metabolizedor Excreted too quickly. Software programscheck molecules for key features(known as “Lipinski’s Rule of Five”)that are associated with favorableADME profiles, such as having five orfewer hydrogen bond donors and amolecular weight below 500.With enough computing power, scientistscan also virtually screen a candidatecompound against a large panel ofproteins from the body, to make sure thecompound will not cross react with otherenzymes or receptors to cause side effects.To ensure that molecules identifiedin the computer will have real-worldFor the field to progress, says Anthony Nicholls, the currentsoftware needs to be more closely scrutinized—using prospectivestudies that directly compare the impact of computer-aidedmethods with more traditional drug design approaches.www.biomedicalcomputationreview.orgSummer 2007 BIOMEDICAL COMPUTATION REVIEW 29

DOCK THIS:Drug Design Feeds Drug Development“I think in the next seven to ten years, with the computationalpower that’s coming on line here pretty soon and the steadydevelopment in algorithms, computer-aided design is going tomake a huge difference,” says Richard Casey.value, computational scientists benefitfrom working closely with medicinalchemists during lead identification andoptimization.“Medicinal chemists would tell youthat there’s lots of intuition involved, soit’s not all computational,” says HansWolters, PhD, associate director ofinformatics at XDx, Inc. For example,he says that as computer scientistsbecame more involved in making drugs,the molecular weight of candidate compoundsbegan to creep up precipitously—tosizes that would not be easilyabsorbed by the human body.Medicinal chemists help recognize thistype of problem early in the process.DEBATING THE IMPACTIn the past two decades, althoughcomputer-aided drug design hasbecome an integral part of drug discovery,some remain skeptical as towhether these methods are deliveringon their promise. The productivity ofthe pharmaceutical industry has actuallydeclined in the past decade (TheFDA approved 58 drugs from 2002 to2004 compared with 110 from 1994 to1996, according to the Tufts Centerfor Drug Development.) Though thisis likely due to many factors—in particular,tightening safety standards andthe enormous cost and time of clinicaltrials—the trend has left some wonderingwhether large investments in technology,including computer-aideddrug design, are paying significantdividends.Many modeling programs are unreliable,and they are not making a big differencein the real world, cautionsAnthony Nicholls, President andCEO of OpenEye Scientific Software,which develops software for computeraideddrug design. “It’s all done onfaith. It’s all done on the idea that ‘oh,we’re using computers, so it must bebetter,’” he says. “I think a lot of peopleare fooling themselves.” He believesthat, for the field to progress, the currentsoftware needs to be more closelyscrutinized—using prospective studiesthat directly compare the impact ofcomputer-aided methods with more traditionaldrug design approaches.Other scientists agree that the algorithmsare still being refined, but have amore optimistic outlook. They say thatprogress is steady and that computeraideddesign is already having animpact. Klaus Klumpp, PhD, an associatedirector at Roche (who wasinvolved in the development of the HIVprotease inhibitor saquinavir), points toa suite of emerging drugs for hepatitis Cvirus (HCV) as a case in point.HCV was discovered in 1989 andthe virus was difficult to grow, so structuralinformation for HCV polymeraseand HCV protease became available relativelylate—in the mid-to-late 1990s. Bythis time, computer-aided drug designwas well integrated into big pharmaceuticalcompanies. Several companiesquickly identified binding sites anddesigned inhibitors, many of which arenow in early clinical trials. “It is expectedto completely change the treatmentparadigm for HCV infected patients,”Klumpp says.Richard Casey, PhD, founder andchief scientific officer of RMCBiosciences, Inc., has also witnessed thedramatic effect that computers can haveon drug design. His company providescomputer-aided drug design services forsmall and mid-size pharmaceutical companies,which often lack in-house teams.Recently, he made 3-D models andperformed in silico docking studies for amid-size pharmaceutical company thathad identified active lead compounds buthad no understanding of how they werebinding the target, an RNA synthetase.“When they saw this for the firsttime, it was the ‘aha’ effect: So that’swhy this compound has high activityand this compound does not. It was areal eye-opener for them,” Casey says.“I think in the next seven toten years, with the computationalpower that’s coming on line herepretty soon and the steady developmentin algorithms, computeraideddesign is going to make a hugedifference.” ■30 BIOMEDICAL COMPUTATION REVIEW Summer 2007 www.biomedicalcomputationreview.org