here - Biomedical Computation Review

here - Biomedical Computation Review

D I V E R S E D I S C I P L I N E S , O N E C O M M U N I T YREVIEWPackingIt All InCurricula forbiomedicalcomputationPLUS:The dawn ofBRAIN–MACHINEINTERFACESBrain implants aregiving hope to thedisabled andrevolutionizingneuroscienceFall 2005

contentsContentsFall 2005714FEATURESPacking It All In:Curricula forBiomedicalComputingBY SHAWNE NEEPERThe Dawn ofBrain-MachineInterfacesBrain implants aregiving hope to thedisabled andrevolutionizingneuroscienceBY KRISTIN COBB, PhDDEPARTMENTS1 From the Editor:Welcome BackBY DAVID PAIK, PhD2 News Bytes• Bacteria with Byte• An Unfolding Story• From Rods and Joints to Interactive Handheld Molecules• PLoS Computational Biology Launched• Simulating Faulty Folding: A Theoretical Model of Prion Propagation• Reliable Models Now Available22 Editor’s Picks BY DAVID PAIK, PhD24 Under the Hood: Twin Curses Plague Biomedical Data AnalysisBY RAY SOMORJAI, PhD26 Seeing Science: Art That’s a BlastBY KATHARINE MILLEROn the Cover: Human Profile © Photographer: Alex Puentes, and Sperm © Photographer: Daniela Spyropoulou, Blood Cells © Photographer: Monika Wisniewska, © Photographer: Dimitrios Kessaris, of Enzyme from M. Tuberculosi, Credit: Mycobacterium Tuberculosis CenterCover art by Rachel C. Jones of Affiliated Design.Fall 2005Volume 1, Issue 2ISSN 1557-3192Executive EditorDavid Paik, PhDManaging EditorKatharine MillerScience WritersKristin Cobb, PhDKatharine MillerShawne NeeperCommunity ContributorsRay Somorjai, PhDLayout and DesignAffiliated DesignPrintingAdvanced PrintingEditorial Advisory BoardRuss Altman, MD, PhDValerie Daggett, PhDScott Delp, PhDEric Jakobsson, PhDRon Kikinis, MDIsaac Kohane, MD, PhDTamar Schlick, PhDJeanette Schmidt, PhDMichael ShermanArthur Toga, PhDShoshana Wodak, PhDJohn Wooley, PhD,REVIEWFor general inquiries,subscriptions, or letters to the editor,visit our website atwww.biomedicalcomputationreview.orgOfficeBiomedical Computation ReviewStanford University318 Campus DriveClark Center Room S170Stanford, CA 94305-5444Biomedical Computation Review is publishedquarterly as part of the disseminationefforts of Simbios, one of the NIH NationalCenters for Biomedical Computation (NCBC).The NIH program and science officers for thegrant are:Peter Lyster, PhD (NIGMS)Peter Highnam, PhD (NCRR)Jennifer Couch, PhD (NCI)Semahat Demir, PhD (NSF)Charles Friedman, PhD (NLM)Jennie Larkin, PhD (NHLBI)Jerry Li, MD, PhD (NIGMS)Wen Masters, PhD (NSF)Richard Morris, PhD (NIAID)Grace Peng, PhD (NIBIB)David Thomassen (DOE)Ronald J. White, PhD (NASA/USRA)BIOMEDICAL COMPUTATION REVIEW Fall 2005

from the editorFrom theEditorDAVID PAIK, PhD, EXECUTIVE EDITORWelcome BackOne of the main goals of this magazine isto create and foster a sense of communityamong the diverse disciplines thatmake up biomedical computation (hence ourtagline: diverse disciplines, one community).More than 2,500 biomedical computationresearchers received the first issue of BiomedicalComputation Review in print. You come from all50 states and many overseas nations. Your fieldsof study are varied—from computer science tostatistics to biochemistry to clinical medicineand beyond. In this issue, we cover topics in specificvertical interest areas as well as cross-cuttingissues that affect us all.VERTICAL FOCUSSome of the topics that BCR covers are specificto individual disciplines. Many excitingthings are happening in the realm too small forthe human eye. In this issue, Katharine Millerand Kristin Cobb uncover a variety of hot newstopics in the News Bytes column, includingcomputer simulation of nucleosomal array folding,chromatin folding and bacterial behavior, atheoretical model of prion propagation, augmentedreality for molecular structure and thelaunch of PLoS Computational Biology.In the macroscopic arena, traditionalhuman-computer interfaces consist of keyboardsand mice. But in one of the feature articles,Kristin Cobb reports on the ultimate humancomputerinterface: direct neuronal connectionsvia tiny electrode arrays. Such technologyoffers the tantalizing possibility of restoringmovement to quadriplegics, and unlocking peoplewith ALS (Lou Gehrig’s disease) from theirlives of isolation.There is a science to engineering high qualitysoftware and this is an aspect of biomedicalcomputing that cannot be ignored. InEditor’s Picks, we discuss a variety of mediaformats, including anonline essay, a book, anda development environmentthat are all apropos to software engineering.“Beauty is truth, truth beauty,” wrote Keats.He probably wasn’t talking about biomedicalcomputing, but we think it still applies. Onthe back cover, art and sciencecome face to face;BLAST local sequence alignmentsare used to create apostmodern artwork called“Ecce Homology.”CROSS-CUTTING ISSUESAs in any large community,there are cross-cuttingissues that are of potentialinterest to all. Education isone of those. In the other featurearticle, Shawne Neepertakes an in depth look atinterdisciplinary training inbiomedical computation.If less is more, then conversely,more is less. In Underthe Hood, community contributorRay Somorjai gives abrief tutorial on how the related curses of dimensionalityand data sparsity affect the computationalanalyses of many varieties of biomedical data.COMMUNITY CONTRIBUTIONAs a service to this community, we’re workinghard to provide the most interesting and up-todatecoverage of biomedical computing that wecan. But we need your feedback and your help.Think of this magazine as ours (as in yours andmine); let us know what you feel strongly about;and consider submitting contributions. Contact usat ■“Beauty is truth,truth beauty,”wrote Keats.He probably wasn’ttalking aboutbiomedicalcomputing, butwe think itstill applies.www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 1

NewsBytesInteractiveHandheld MoleculesThirty years ago, molecular biologistsroutinely constructed proteinmodels out of brass rods (“Kendrewmodels”). In recent years, researchersput away such tinker toys and turned tocomputer graphics.But now scientists at The ScrippsResearch Institute are combining thetwo mediums. They “print” threedimensionalmodels of biological moleculesthat, when held and manipulated,interact with the computer that printedthem. The work was published in theMarch issue of Structure.“Everyone has a gut feeling thatthere’s something different about holdingan object versus looking at it on thescreen,” says Art Olson, PhD, professorof molecular biology and director of theMolecular Graphics Laboratory at TheScripps Research Institute. “Butbecause these models are essentiallycomputer output, they have a specialrelationship to the data in the computerthat actually made them.”Olson and his colleagues generatehandheld molecules with 3D fabricatingprinters that can make solid objects outof layers of plaster or plastic. Then, as aperson turns or twists the object, a digitalvideo camera tracks its movements.The computer displays these manipulationsas well as additional informationabout the molecule in what is known as“augmented reality.” For example, as aperson moves two physical moleculestoward one another, the screen mightshow how the electrostatic fields andelectrical potentials change.Why bother with physical models atall? “If you can print out a custom interfacethat’seasy to handleand willaddress whateverproblemyou’re interested in, then why not try todo that?” Olson says. “You have moretools and more cues if you have the analogphysical object.”Physical models with augmentedreality have an advantage over purecomputer models because they’re moreeasily manipulated, Olson says. “It’seasy to tie a knot in a string with yourhands. It’s much harder on the screen.”“It’s easy to tie a knot in a stringwith your hands,” says Olson. “It’smuch harder on the screen.”Given a flexible model of a protein, aresearcher can pick apart the end terminusand see how it might interact iflain against a different part of itself.“Doing that with a mouse would be relativelydifficult.”Above: Computer augmentation of two subunits of the SODdimer which are tracked and manipulated independently. Theelectrostatic field is shown with small arrows that point alongthe local field vectors (they appear as small as dots in thispicture), and the potential is shown with volume renderedclouds, with positive in blue and negative in red. At Right: Aphysical ball and stick model of an amino acid is augmentedby computer graphics showing the spacefilling model superimposed.Courtesy of Art Olson, Molecular GraphicsLaboratory, The Scripps Research Institute.2 BIOMEDICAL COMPUTATION REVIEW Fall 2005

The physical models might alsoprove valuable as talking devices whenstructural biologists collaborate withscientists who don’t routinely thinkabout structure, Olson says. Andadding augmented reality to physicalmodels may prove helpful in explainingcomplex concepts to students. In earlytests, one thing is for sure, says Olson,“The students like it better.”Bacteria with ByteWhen a bacterium swims towardfood, it follows a chaotic path, alternatingbetween spinning randomly anddriving forward, or ‘tumbling’ and ‘running.’Computer scientists at theUniversity of Chicago have now createda virtual colony of E. coli bacteria—complete with digital receptors,motors, and signaling pathways—thatrun and tumble just like real bacteria.The simulation program, AgentCell,is the first to model a biochemical networkat the molecular, single cell, andpopulation levels simultaneously. Bydoing so, it might provide a frameworkfor modeling other biological systems,including cancer and antibacterialresistance. AgentCell was introduced inthe June 1 issue of Bioinformatics.www.biomedicalcomputationreview.orgDigitalE. coli swimrandomly ina nutrientfreemedium(green cells)or up agradientof nutrient(red cells).Solid redand greenlines indicateaveragepositionof eachpopulation.Courtesyof ThierryEmonet.“We usedAgentCell’schemotaxis network, but you can usethe program for any kind of networkyou want,” says Thierry Emonet, PhD,a research scientist at the University ofChicago and lead author on the paper.AgentCell uses agent-based simulation,a type of software developed tomodel social behavior, such as thestock market. An agent is a softwareobject that makes completelyautonomous decisions. InAgentCell, each single-celledbacterium decides to run ortumble based on input fromthe virtual environment andfluctuating intracellular signals.The program models thebehavior of thousands of bacteriaacting independently.Future versions will allow thebacteria to interact.The researchers used bacterialchemotaxis as a test-bedfor AgentCell because it is oneof the best characterized systemsin biology. An E. coli bacteriumswims toward nutrients and away frompoisons by alternating the rotation ofits flagella: counterclockwise motioncauses flagella to bundle into a tightpropeller (running); clockwise motioncauses the flagella to fly apart (tum-bling). Chemical signals in the bacteriacontrol the switch between run andtumble, but the mechanism is noisy:the frequency of switching is highlyvariable between two genetically identicalcells in the same environment.Using AgentCell, Emonet hopes tobetter understand how cells make decisionsin the face of such variability.Chemotaxis is super-simple decisionmaking;a more complicated decisionfor a cell is whether or not to divide,Emonet says. When customized,AgentCell could be used to study howcell division goes awry in cancer.Computer simulations can testcompeting theoretical models andguide future lab experiments. In thelab, it takes months to grow cells witha mutated protein. In the computer, ittakes just a quick and elegant changeof the code.AgentCell will soon be available asopen access code on the “We’d love to havepeople grabbing the code and addingmodules, adapting it to their ownneeds,” Emonet says.“Because of its modular architecture,the system readily integrates preexistingsimulators and algorithms withvery little development overhead,” saysIn the lab, it takes monthsto grow cells with a mutatedprotein. In the computer,it takes just a quick andelegant change of the code.Tom Schmizu, a post-doctoral fellowat Harvard University. For example,you could easily swap in your ownfavorite model of intracellular signalingor of receptor binding intoAgentCell, and then test it againstdata from the real world. >Fall 2005 BIOMEDICAL COMPUTATION REVIEW 3

NewsBytesUsing computer simulations, agroup of theoretical chemists atWarsaw University—Malolepsza,Boniecki, Kolinski and Piela—managedto get a glimpse of how suchmisfolded proteins—called prions—propagate. Recently, they designed aprotein that, in computer simulations,induces other proteins to misfold.The work was published inProceedings of the National Academyof Sciences in May.Edyta Malolepsza, a graduate studentto what develops in prion disease.Malolepsza cannot yet explain whyone of the proteins could propagatemisfolding while the related sequencescould not. Figuring this out may yieldclues about how to inhibit or reverseprion disease.“Only one among the studiedsequences exhibits the ability to induceprion disease,” says Malolepsza. Andjust a few amino acids—sometimes onlyone—made the difference between theprotein that acted like a prion and theJust a few amino acids made thedifference between the protein that actedlike a prion and the 13 others that didn’t.A model of prion disease propagation. Thecorrect (helical) form of a protein misfoldsto a beta-structure in the presence of thestiff misfolded beta-structure form.Courtesy of Edyta Malolepsza.SimulatingFaulty Folding:A Theoretical Modelof Prion PropagationInside a live cell, strings of aminoacids instantaneously fold into proteinswith very specific shapes. Typically, noharm is done if a protein somehow foldsinto an unconventional configuration.But, very rarely, a misfolded protein willinduce others to unfurl and misfold aswell, with disastrous consequences: thenonconformists glom together, causingdiseases such as bovine spongiformencephalopathy (BSE or mad cow),Creutzfeldt-Jakob, and Alzheimer’s.involved in the work, says she hopes itwill help advance our understanding ofprion disease. “This is a small modelwith a protein designed by us, not bynature,” she says. “But because we useda very realistic force field, a real proteincould behave similarly.”Malolepsza’s work involved two primarysteps: designing proteinsequences that might have a propensityto misfold, and simulating what happenswhen they interact. For thedesign process, she used trial and errorto identify a set of 32-amino-acidchains that met specific criteria: theywould naturally fold into a bundle oftwo alpha-helices, but, at only a slightlyhigher energy level, could also forma beta-sheet.After selecting 14 appropriatesequences, Malolepsza began her simulations.Alone, each peptide folded tothe native alpha-helical shape at a varietyof temperature ranges. However,when allowed to interact with a frozenbeta-sheet version of itself, one of thesequences (regardless of its startingconformation) unfolded and thenrefolded to a beta-sheet shape. It thenformed a dimer with the pre-existingbeta-sheet. In addition, allowing onefrozen beta-sheet molecule to interactwith two alpha-helices produced abeta-trimer—a larger aggregate similar13 others that didn’t. “Maybe a mutationoccurs that allows propagation ofthe amyloid aggregations seen in priondisease,” says Malolepsza.Malolepsza is hoping to simulateactual prion proteins soon. It’s a morecomplex task because prions are bigger—thefragment needed for simulationscontains about 100 amino acids.“We need a faster simulating program,”she says. “I hope that we will eventuallyhave another paper with a more completeanswer as to how prions work.”An Unfolding StoryTo fit an organism’s DNA into a singlecell, it has to be tightly compacted,first wound around proteins to formchromatin fibers, then further coiledinto chromosomes. Computer simulationsby scientists at New YorkUniversity (NYU) have now provideda better understanding of how this foldingoccurs. The results appeared in theJune 7 issue of the Proceedings of theNational Academy of Sciences.“It’s very important to understandhow chromatin folds and unfolds,” saysTamar Schlick, PhD, professor ofchemistry, mathematics, and computerscience at New York University andsenior author on the paper. Chromatinfolding is directly involved in gene4 BIOMEDICAL COMPUTATION REVIEW Fall 2005

PLoS ComputationalBiology LaunchedIn June, the Public Library ofScience (PLoS) teamed up with theInternational Society forComputational Biology to publishthe first issue of PLoSComputational Biology. Editorshope that bringing the best worktogether in one place will boost perceptionsof the field’s importance.“There’s no journal that’sdevoted to our understanding ofliving systems using computationalbiology,” says Philip Bourne, PhD,professor of pharmacology at theUniversity of California, SanDiego, and editor-in-chief of thenew journal. Such papers getpeppered all over PNAS, theJournal of Molecular Biology,Science, Nature, and Cell, he says.“That doesn’t help solidify thevalue of computation to biology.”By focusing on biologicalapplications, the new journalshouldn’t infringe on the territoryalready covered by Bioinformaticsand The Journal of ComputationalBiology, which are oriented moretoward methods and algorithmdevelopment. “The future of thefield is really with people whodevelop and then apply methodseither in in silico labs or in conjunctionwith wet labs,” Bourne says.Rather than focusing on computationalmolecular biology, thejournal is trying to broaden theperspective. “We’re getting verydiverse papers with a computationalthread,” Bourne says. “Thatincludes papers on such topics ascomputational neurobiology,population genetics, andcomputational ecology.”Bourne understands that whenpeople have produced an incrediblepiece of work, they might still tryfor Science or Nature. “But,” hesays, “over time we are trying veryhard to make PLoS ComputationalBiology the place where they sendtheir best quality work.”expression and silencing: chromatin—a complex of DNA and specialized proteins—mustunwind so that the cellularmachinery can access the DNA andbegin copying or transcribing thegenetic information into proteins.A stretched-out chromatin fiberlooks like “beads on a string.” TheDNA is wound around repeating 8-proteincomplexes at approximately regularintervals; each DNA/protein “bead”is called a nucleosome. Scientistsalready knew that chromatin unfolds inlow-salt solutions and folds in high-saltsolutions, such as found in cells. Butthey couldn’t distinguish between fourpossible folding structures (perpendicularand parallel zig-zag, and perpendicularand parallel solenoid), until now.The scientists at NYU modeled thefolding of a 12-nucleosome fragment ofchromatin using what they believe isthe highest-resolution simulation ofchromatin folding to date. Chromatin istoo large and complex to model atomby-atomwith today’s computing power.But modeling at the level of macromolecules(proteins and DNA) is too crudeto give a realistic picture. So, NYU scientistscompromised: Using structuralexperimental information about eachnucleosome and the electrostatic forcesassociated with each atom, they built arealistic mechanical model containingessential features of the system whileapproximating others. They modeledthe key positive and negative chargesfound on the amino acids andnucleotides, without explicitly modelingevery atom. Chromatin foldsaccording to the attraction and repulsionof these charged particles witheach other and with the salt solution.“This allows us to do long-timesimulations of the complex systemusing what is a very realistic model ofA 12-nucleosome array adopts extended beads-on-a-string conformations in a lowsalt solution (outer ring), while it compacts at midlevel salt concentrations (middlering) and folds into irregular zig-zag structures at high salt concentrations (inner ring).Courtesy of Tamar Schlick.www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 5

NewsByteswhat the nucleosome core wouldlook like.” Schlick says.Regardless of which of the four foldingmodels they started their simulationwith, they found that their virtualchromatin always folded into an irregularzig-zag conformation after enoughcomputational steps. They also pinpointedthe key electrostatic attractionsand repulsions that drive chromatinfolding and unfolding.“This is not the first attempt tomodel the chromatin fiber, but this onemakes the fewest artificial assumptions,”says Sergei Grigoryev, PhD,assistant professor of biochemistry andmolecular biology at Penn StateUniversity College of Medicine.Their findings agree with the experimentaldata he has collected on chromatinfolding using electron microscopy.“I really admire their paper,” he says.“For the first time it produced a nucleosomearray model that really matchesbiological observations.”A graphical depiction of the private BioModels submission (1), checking (2),and annotation (3) steps (above) and the public access steps (below).Courtesy of Nicolas Le Novere.Reliable ModelsNow AvailableAs systems biologists develop modelsthat attempt to simulate life, theyneed a good way to make them accessibleto others as well as a good way toaccess other peoples’ models—and toknow they can be trusted to work. Aninternational collaboration known asBioModels intends to provide just that;in April they released an initial set offully annotated models for public use.“We are storing quantitative, peerreviewedmodels so that people canuse them,” says Nicolas Le Novere,PhD, a computational neurobiologistwith the European BioinformaticsInstitute (EBI) in the UnitedKingdom. “We want it to be a kind ofgolden resource.” BioModels is theresult of a collaboration led by EBIand the SBML Team, an internationalgroup that develops open-source standardsto describe biological systems.The project staff only accepts modelsthat have been published in peerreviewedliterature. Curators thencheck to make sure that, when downloadedand run in the appropriate simulationsoftware, the model will dowhat it’s supposed to do. Next, annotatorsadd model descriptions and crosslinksto related modelsand papers. At thatpoint, the model isreleased for public use.The systems biologycommunity is wageringthat this collectionof models willprove extremely valuable.According to aneditorial in Nature,“It is hoped thatBioModels will formthe basis of a universallyaccepted repositorythat can do forsystems biology whatGenBank and theProtein Data Bankhave done for geneticsand structuralbiology.” Nature 435,1 (5 May 2005)The majority of early submissionsto the database deal with signalingpathways or metabolic networks, butthey are quantitative and dynamicmodels—not just pathways. “You canimport these models into a simulator,click ‘run,’ and see things happen, see“We wantit to be akind of goldenresource,” saysPhilip La Novereof theBioModelsproject.values updated,” says Le Novere.Formalized, realistic models of subcellularparts or even muscles can alsobe stored in BioModels. And althoughmodels of that typehaven’t arrived yet,Le Novere says theproject already has abacklog of submissions.“We have somany good modelsarriving that we haveto prioritize.”BioModels initialusers are primarily thepeople who’ve createdthe models, says LeNovere, but he anticipatesthat will soonchange. The siteshould prove extremelyvaluable to experimentalbiologists whowant to have an ideaof how a system worksbefore designing anexperiment. Andpharmaceutical companies could turnto it as well, in order to test the likelyeffect of enhancing or inhibiting a moleculeor doing things that affect severalparts of a network at the same time.For more information, visit ■6 BIOMEDICAL COMPUTATION REVIEW Fall 2005

PACKING ITALL IN:Curricula forBiomedicalComputingThe last decade saw a proliferationof training programs atthe intersection of life scienceand computation,with more than 60 newdegree and certificate programs launchedin the U.S. alone—and a similar number worldwide.Most appeared within the last five years. 1The number and variety of programs aregrowing, in step with the complexity ofmodern biomedical challenges.It’s the evolution of biological researchthat’s driving educational change. “We’renot just sequencing DNA or looking at agap junction,” says John Wooley, PhD,Associate Vice Chancellor of Research atUC San Diego and an active advocatefor education in biomedical computation.“We’re putting together thetotal cardiovascular system fromthe heart to the microcirculation, orthe brain from neuron junctions allthe way up to cognition.” This workrequires interdisciplinary education,says Wooley, who is also Senior Fellowof the San Diego Supercomputer Center.“It’s going to be the next generationwho actually accomplish these goals—studentswho understand both biology andcomputing, and can listen to experimentalbiologists and mathematicians.” >BY SHAWNE NEEPERwww.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 7

To train that new generation,recent reports fromthe NSF, NIH andMathematical Associationof America proposechanges across the educational pipeline,from K-12 through postdoctoral. 2,3,4,5First, they suggest, introduce undergraduatecoursework to make all biologistsmore conversant in math—not just calculus,but discrete mathematics, matricesand probability. Further, add undergraduatebiology requirements in algorithmand software design. Finally, createcross-disciplinary graduate and postdoctoralprograms to train future theoristsand tool builders in math, computerscience and life sciences.HOW MUCH CANONE BRAIN DO?With so much ground to cover, educatorsat all levels face a tug-of-warbetween breadth and depth.Quantitative challenges in biomedicinedraw on a daunting array of computationalskills—across mathematics,statistics and computer science—aswell as various strengths in physics,chemistry and engineering. And anytopic in biology alone could consume alifetime’s study.There's a spectrum of opinionsabout how much one brain can do. “Ithink depth is a critical issue,” saysKeith Elliston, PhD, co-founder, presidentand CEO of Genstruct, Inc. “Ifyou try to train in two disciplines, youend up with people who are pretty shallowin both.” Genstruct needs expertswith advanced degrees in molecularbiology, a disease area, or computer sci-8 BIOMEDICAL COMPUTATION REVIEW Fall 2005

© PHOTOS.COMEducators atall levels face atug-of-war betweenbreadth and depth.ence, Elliston says. Genstruct trains thebiologists in-house to build or applydrug and disease models. Computer scientistsbuild the underlying software. “Icertainly see providing people withsome exposure and cross-training, but Ithink you need specialists first for thatto have real value,” Elliston says.What’s the best balance of depthand breadth? Genstruct’s approachillustrates one end of the spectrum:computational biology as collaborationbetween experts in the life sciences andcomputation, with enough cross-trainingto create common ground. Theopposite extreme falls just short ofrequiring a quadruple major in computerscience, math, statistics and biology.New educational offerings span a continuumbetween Genstruct’s positionand broad interdisciplinary approaches.Interdisciplinary training programsmust evaluate which skills are essential,and determine where program boundariesshould lie. As programs develop,educators are defining new fields andredefining traditional study.CROSSING DISCIPLINESOne way to ensure depth is to trainbiomedical degree holders in a quantitativefield, or vice-versa. For example,in the 1990s, progress in genomicsmotivated the Alfred P. SloanFoundation to fund postdoctoral grantsthat cross-trained quantitative scientistsin molecular biology. Inspirationfor the program came from leadingmolecular biologists, says MichaelTeitelbaum PhD, program director atthe Sloan Foundation. “They were sayingthings like ‘I’m computer literateand numerate, but the volume of data isexpanding geometrically, and I’m not ahigh-powered computer scientist. Idon’t have the skills I’d need to dealwith such large data sets.’”At the same time, people with PhDsin computationally intense fields suchas physics and mathematics faced a difficultjob market. So the SloanFoundation, joined by the Departmentof Energy, funded two-year postdoctoralfellowships for physicists, chemists,engineers and mathematicians to workin molecular biology laboratories.From 1996 to 2003, more than 60awards supported cross-training atinstitutions across the U.S.“Our expectations of the rate ofgrowth were too conservative,”Teitelbaum says. “We didn’t anticipatethe almost vertical take-off thatoccurred in the 1990s. These peoplewere in high demand when they finishedtheir postdocs.”Experienced molecular biologistsalso need continuing education tolearn how to use new bioinformaticstools as they’re developed. For example,the NIH’s National Center forBiotechnology Information, whichproduces PubMed and such tools as theBasic Alignment Search Tool amongothers, teaches courses ranging fromgenomics search strategy to scripting ofcustom BLAST searches. Courseattendees include roughly one-thirdeach of NIH, academic and industryusers, says David Wheeler, PhD, headof user services for the NCBI.CROSS-TRAININGFOR INDUSTRYFor people who already have thedepth of at least one science degree,there exists another form of crosstrainingin bioinformatics: the interdisciplinaryscience master’s degree,geared toward people who are alreadyin the workforce.“Companies need people who canwork in teams with molecular biologists,chemists, and other specialists onresearch and development projects thatneed computational molecular biologyskills,” Teitelbaum says.The Sloan Foundation funds interdisciplinaryscience master’s degreeprograms, 6 many of which focus onbioinformatics and computational biology.These programs train degree holdersin the sciences, mathematics, engineeringor a health discipline to bringcomplementary skills into the workplace.The two-year courses are tightlyfocused on applications and ofteninclude projects with industry partners.“A significant proportion of our studentsare people at the prime of theircareers,” says Betty Cheng, PhD, associatedirector of Stanford University’sBiomedical Informatics TrainingProgram. “They use this program todevelop new directions within the corporatesetting.” One such student wasresponsible for starting a clinical informaticsprogram in his healthcareorganization. Another is applying hisnew skills to create multimedia healthinformation technology for patientswith disabilities.The Keck Graduate Institute (KGI)www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 9

launched its professional master’s programfour years ago. “[This] is a professionaldegree, on a different track thanresearch,” says Gregory Dewey, PhD,vice president for academic affairs anddean of faculty at KGI.The program was founded on thepremise that industry needs scientificallyand computationally astute teammembers who aren’t theorists or benchscientists. KGI graduates find positionsin bioinformatics, product development,project management, regulatory andclinical affairs, and sales and marketing—primarilywith biotech, pharmaceuticaland medical device companies.INTEGRATED STUDYRather than teaching computationto life scientists, or vice-versa, interdisciplinaryprograms increasingly teachboth at once. They aim for breadth, butalso to define new fields in bioinformaticsand computational biology.Concurrent training in life scienceand computation began decades agowith an emphasis on clinical informatics,and grew to embrace genomics,proteomics, and more.Interdisciplinary programs have proliferatedwith the help of funding organizationslike the NIH’s National Libraryof Medicine (NLM). In 2001, the NLMlaunched a program that now supports18 medical informatics research trainingprograms. The programs offer integratedgraduate degrees and promoteinterdisciplinary research. 7“We’ve accepted a potpourri of programs,”says Milt Corn, MD, associatedirector, Division of ExtramuralPrograms for NLM. “The field is sonew. We do not prescribe a specific curriculumor organizational structure.”Recently, the NLM inspected all18 of its programs and found that universitieshave shaped them in a varietyof ways, Corn says. In some cases,programs are divided by topic area.Other Universities are trying to putcomputation under one roof, with theidea that different topics sharemethodologies. “Which of these modelswill have ascendancy is not for meto say,” Corn says. “Given how ourcountry works, they’ll all probablycoexist with one another.”“We envisiona change in thecurrent approachto graduate schoolwhere most of thestudents take thesame courses, joina lab, and work onsome aspect of theadvisor’s project,”says Pamela Silver,director of thenew systemsbiology program atHarvard University.Many life scientists in differentfields use similar computationalapproaches, says Atul Butte, MD, PhD,who recently obtained his doctorate inintegrative biology at the Harvard-MITDivision of Health Sciences andTechnology. “So I’m more optimisticthan many that there are ways to trainthese folks at the same time.”As an example, Butte points tomachine learning. “The computer learnspatterns in data, to distinguish x and y,”he says. “For a medical researcher, x andy could indicate disease state; for a structuralbiologist, whether a molecule goeshere versus there.”Similarly, genomic, systems, andmedical researchers often require relatedmethods of classification and treebuilding, or may share approaches tomodel building. Butte—who joinedthe medical informatics faculty atStanford University this year—arguesthat these common methods can formthe backbone of a core curriculum incomputational biology.BEYOND BIOINFORMATICSBiomedical computation now reacheswell beyond the boundaries of bioinformatics,as reflected in emerging graduateprograms in computational andsystems biology. These programs followthe lead of pre-existing neuroscienceand ecology programs by teaching studentsto use computational methods tomodel complex biological problems. Forexample, they might develop algorithmsfor protein folding, blood flowthrough vessels, neuron action potentials,or tumor growth. Informaticsmight play a part in such programs, butit is not necessarily the focus.10 BIOMEDICAL COMPUTATION REVIEW Fall 2005

Harvard, Yale and theMassachusetts Instituteof Technology (MIT)have all created newPhD programs in the lasttwo years. Harvard’s new SystemsBiology PhD program gets underwaythis fall with a surprisingly flexible curriculum.It has only two requiredcourses: a survey of faculty research,and an exploration of potential newapplications of computation in biomedicine.Unlike most bioinformaticsdegrees, it has no course requirementsin molecular biology.“We envision a change in the currentapproach to graduate school wheremost of the students take the samecourses, join a lab, and work on someaspect of the advisor’s project,” saysprogram director Pamela Silver, PhD,professor of biological chemistry andThe MIT curriculum isdesigned around what facultycall the “four Ms”: Measurement,mining, modeling and manipulation.© PHOTOS.COMmolecular pharmacology at Harvard.“We would like to get away from thatmodel and empower students more.”Students will have two facultyadvisors in their first year: one inphysical or quantitative science,another in biology. But they areencouraged to propose inter-lab collaborationsfor their thesis work andmay select an advisor from anywherein the University. The program wasannounced in fall, 2004. ByDecember, about 100 applications hadarrived from students of computer science,biology, chemistry, engineering,medicine, and more.Founded in 2004, MIT’s PhD programin computationaland systems biology offerssimilar flexibility to theHarvard program. Thecurriculum is designedaround what faculty call the “four Ms”:Measurement, mining, modeling andmanipulation. Classes stress real-worldexamples to explore the integration ofthese approaches in modern research.“Research these days is multiinvestigatorand multi-site,” saysBruce Tidor, PhD, professor of bioengineeringand computer science atMIT and director of the new program.“Learning how to work in thatenvironment is crucial. Many of ourstudents will have more than onesupervisor, and thesis projects willintegrate their different expertise. Bylearning ways of solving problems ineach of their research groups, our studentswill be in a position tobridge disciplines.”Many of the new PhD programsoverlap with researchareas in existing programs atthe same institutions. Forexample, students interestedin PhD study in computationalbiology and bioinformaticsat Columbia University canapply to a new interdepartmentalprogram, or pursuesimilar research throughexisting PhD programs in biochemistry,molecular biophysics,biomedical informatics,biological sciences, computerscience, electrical engineeringor applied mathematicsand applied physics. Sowhy have both options? Thedifference is in the design: thenew PhD programs, such asthe one at Harvard, span theother departments, and more.“It gives us a degree of flexibilityyou don’t have in a traditionaldepartment setting,”Silver says. “It allows us todefine the field as we go.”MIT’s program favors applicantswith dual backgroundsin the life sciences and a quantitativediscipline, already preparedfor immersion in cross-www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 11

disciplinary work. And more studentsare now coming to graduate schoolwith that kind of background.STARTING YOUNGERSome in the field believe that amore thorough integration of computationand biology should beginbefore graduate school. But theconflict of breadth and depth intensifiesin younger students whohaven’t yet had time for depth.How can educators include morequantitative work in an alreadycrowded biology curriculum?Several universities have alreadylaunched undergraduate degrees inbioinformatics or systems biology.These programs can be rigorous—oftenthe equivalent of a double major—andfrequently must balance biology andcomputational content.For example, George WashingtonUniversity’s departments of biology andcomputer science offer a dual degree incomputer science and biology, or adegree in computer science with abioinformatics specialization. The latterincludes all typical premedicalcourses in chemistry and physics, a fullcomputer science curriculum, and specialclasses in bioinformatics and computationalbiology. This leaves room foronly two biology classes: introductoryand molecular biology.“It’s going to be difficultto create one unifiedcomputational biologycurriculum,” saysBruno Olshausen, PhD,a professor atthe University ofCalifornia, Davis.The University of California, SantaCruz Department of BiomolecularEngineering offers a BS degree in bioinformaticswith similar requirements,including two biology courses and threeupper-division bioinformatics classes.Most undergraduate degrees narrowthe scope by focusing on bioinformatics.Some venture further, butit’s tough to reach consensus abouthow and what to teach undergraduatesin a broad, new field. “When yousay computational biology, it’s not© PHOTOS.COMjust one thing,” says BrunoOlshausen, PhD, principalinvestigator at the RedwoodNeuroscience Institute andassociate professor of neurobiology,physiology, andbehavior at the University ofCalifornia, Davis.Olshausen and other facultywere asked to design anundergraduate program atUC Davis. “To some extent, acore curriculum is possible;modeling methods are similarenough,” Olshausen says.However, he says, at UCDavis, genomics facultywanted to include courses indatabases, which he wouldnot require in neurosciences.He, on the other hand, mightwant students to take acourse in the theory of computation,which genomics facultymight not agree to. “It’s going to bedifficult to create one unified computationalbiology curriculum.”Stanford’s undergraduate degreein biomedical computation addressesthe issue of breadth and depth directly.Students select a “depth”track in biology or computation,with two corresponding“breadth” tracks in the other.There are four tracks: molecularand cell biology, organand organismal biology,informatics and simulation.So, for example, a studentchoosing depth in organ andorganismal biology wouldstudy both informatics andsimulation approaches tothat area.“An element of the solutionis to not make curriculumfor biomedical computationsimply the sum of the componentfields,” says Scott Delp,PhD, chairman of StanfordUniversity’s Department ofBioengineering and co-directorof Simbios, a National Center forBiomedical Computation. “By definition,then they’re only getting introductorymaterial in each field. By carefullydesigning curricula that teachbiological and physical, math and12 BIOMEDICAL COMPUTATION REVIEW Fall 2005

“By carefully designing curricula that teach biological andphysical, math and computation concepts together you canachieve a much higher degree of efficiency and depth at the sametime,” says Scott Delp, PhD, a professor at Stanford University.computation concepts together youcan achieve a much higher degree ofefficiency and depth at the same time.”Interdisciplinary courses and studymodules can benefit all biology students,argues a recent publication ofthe National Academies Press, calledBio2010. 8 The report includes a longlist of computational conceptsimportant to modern biology, forinclusion in revised undergraduateprograms. This and a more ambitioussuggestion to integrate undergraduatecourses across sciences 9 have metwith mixed reviews. 10,11,12Society for Mathematical Biologypresident Louis Gross, PhD, pointsout the difficulty of fitting separatecourses covering those concepts intoalready busy curricula. 13 However, hecites successes at the University ofTennessee, which introduced essentialmathematics concepts—beyondcalculus and statistics—in aredesigned introductory course forbiologists. The class teaches maththemes through life science examples.He endorses this concepts-wise integrationof quantitative skills throughoutundergraduate biology curricula. 14TOWARDS A CONSENSUSSome educators are trying to developunified recommendations forteaching computational biomedicine.A series of workshops held at nationaland international meetings has builtthe basis for an upcoming report,“Educating Bioinformaticians for theChallenges of 21st Century Biology: Awhite paper for the National ScienceFoundation and the InternationalSociety for Computational Biology.”It will be submitted to the ISCB’seducation committee for comment,and shared with the NSF and appropriateDepartment of Energy andNIH organizations.“We hope to inspire them to furtherdiscussion,” says co-author JohnWooley, who is shepherding the newwhite paper’s growth.Inspired by similar white papers inCanada, Japan and elsewhere, thedocument suggests goals and curriculafor degree and certificate programs inbiomedical computation, and addressesstrategies to better integrate biologicaland quantitative science educationin the U.S.“To make progress in the 21st centurywe need the very powerful approachthat bioinformatics and computationalbiology offer,” Wooley says. “And weneed students who understand bothbiology and computing.”The next decade will reveal howthe interdisciplinary approach faresas more graduates hit the job market.A 1999 survey of 16 schools showsthat many with interdisciplinaryundergraduate and master’s degreesfound positions in software development,pharmaceutical and biotechnologyindustries. Most PhD’s andpostdocs, on the other hand, continuedwith academic careers. As interdisciplinaryeducation evolves, a newgeneration of scientists will emerge.Only time will tell which blends ofcomputation and biomedicine willrise to the fore.REFERENCES1See a partial list. See also M, Arzberger P, Cohen J, Hastings A,Holt R, Morse J, Sumners D, Luthey-Schulten Z(2003). Accelerating mathematical-biologicallinkages: Report of a joint NSF-NIH workshopheld February 12-13, 2003 at the NationalInstitutes of Health. Bethesda, MD.3Hastings A, Arzberger P, Bolker B, Ives T,Johnson N, Palmer M (2002). Quantitative biologyfor the 21st century. Report from NSF fundedworkshops on quantitative environmental andintegrative biology, September 7-9, 2000 andDecember 11-13, 2002, San Diego, California.4Byer B, LeBlanc M (2002). Meeting report:Incorporating genomics research into undergraduatecurricula. Cell Biology Education 1(4).5The Mathematical Association of America,Curriculum Reform and the First Two Years,summary reports in biology and health/life sciences.See Research Council of the NationalAcademies, Division on Earth and Life Studies,Board on Life Sciences (2003). Bio2010:Transforming Undergraduate Education forFuture Research Biologists. Report of theCommittee on Undergraduate Biology Educationto Prepare Research Scientists for the 21stCentury, Performed by the National Academy ofSciences with funding from the NIH and HowardHughes Medical Institute.9Bialek W and Botstein D (2004). Introductoryscience and mathematics education for 21stcenturybiologists. Science 303(5659).10Diniz-Filho J (2004). Back to the future:the 1960s view of statistics and mathematicsfor 21st-century biologists. e-letter responseto Science.11Fambrough D (2003). Review of: Bio2010:Read it! Cell Biology Education 2(2).12Steitz J (2003). Commentary: Bio2010—Newchallenges for biology educators. Cell BiologyEducation 2(2).13Gross, L (2004). Interdisciplinarity and theundergraduate biology curriculum: finding a balance.Cell Biology Education 3(2).14Gross L (2000). Education for a BiocomplexFuture. Science 288(807). ■www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 13

THEDAWNOFBRAIN-MACHINEINTERFACESBrain implants are giving hope to thedisabled and revolutionizing neuroscienceBY KRISTIN COBB, PhD14 BIOMEDICAL COMPUTATION REVIEW Fall 2005

Matthew Nagle can move a cursor on a computerscreen with only the power of histhoughts. It’s a remarkable feat for anyone,but especially momentous for Nagle, who isparalyzed from the neck down.Millions of brain neurons fire when the 25-year old wills hishand to move, but until last year, those messages were strandedshortly after leaving his brain. Now, a pill-sized sensor insertedinto the motor area of Nagle’s brain extracts a tiny fraction ofthese signals and sends them to a computer. The computertranslates the brain’s message, “move arm left” or “move armright,” into cursor movement. Soon, Nagle doesn’t think aboutmoving his arm, but moves the cursor directly.Nagle’s is the first brain-machine interface (BMI) to beplaced inside a human brain. The earliest BMIs were developedin animals, and systems that read brainwaves from the scalprather than from surgical implants have been tested in humansfor several decades.While many hailed the human trial as a milestone, the realwow-factor is in the potential of this nascent technology. Withfurther refinement, a BMI could give Nagle more than limitedcursor movement. By feeding brain signals to muscle electrodes,prosthetic limbs, or an exoskeleton suit, it could somedayrestore motor functions—giving him the ability to pick upan object or even walk.The BMI field is growing rapidly. While Nagle’s BMI takessignals from his brain, other implanted BMIs pass signals intothe brain to restore lost senses such as hearing, touch, or vision.These developments are drawing from and inspiring advances inneuroscience, computer science, and engineering. BMIs are alsogiving scientists a new way to study the inner workings of thebrain as we move, feel, remember, and think. The results arealready changing our understanding of how the brain works. >www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 15

“It surprisesme howquickly cellsadapt tothinking ‘mypurpose in lifeis to move thisdot on thescreen’ asopposed to‘my purpose inlife is to movethis limb,”Heetderkssays.THE LANGUAGE OF THE BRAINReach out and grab an object nearyou. To carry out this seemingly simpletask, your brain has to deal with rapidlychanging information on arm position,arm speed, and hand grasping and liftingforces—all coordinated with changingfeedback from your hand and eyes.Scientists’ limited understanding ofhow this unfolds in thebrain goes somethinglike this: When youmove your hand, thereis a flurry of activity inthe arm area of yourmotor cortex, locatednear the top of yourhead. Millions of neurons“talk” by firingaction potentials, orspikes of electrical currentthat sound like abuzzing or cracklingwhen amplified andplayed through a speaker.The message is containedin the firing rateof individual neurons—100spikes per second means something differentthan 5 spikes per second. And scientistshave discovered that they can “hear” a single neurontalk by placing a tiny electrode right next to itin the brain tissue.In the 1980s, work by ApostolosGeorgopoulos, now at the University ofMinnesota, opened the door for motor BMIs.He implanted a microelectrode array in a monkey’sbrain and recorded single-neuron firingrates as the monkey reached in different directions.Using a simple mapping, he correlatedthe arm’s changing three-dimensional position(x,y,z) to the shifting pattern of neural spiking.“Georgopoulos showed fairly convincinglythat broad populations of cells in the motor cortexparticipate in varying degrees,” says BillHeetderks, MD, PhD, associate director ofExtramural Science Programs at the NationalInstitute of Biomedical Imaging andBioengineering (NIBIB). That’s good news forbuilding a motor BMI, Heetderks explains. Togenerate precise control signals you have torecord the firings of hundreds of neurons in thebrain. But you don’t have to be too strict aboutTHE NEURAL CODE Upper panel: Each neuroninvolved in arm movement is broadly “tuned” to aparticular direction. This neuron’s firing ratechanges as a monkey reaches for different flashingtargets (A-G) on a touch screen, firing most rapidlywhen the monkey reaches for E (up and right).Lower panel: Individual neurons are tuned to differentdirections. For example, when a monkeyreaches for target E, particular neurons fire rapidlyand others fire slowly. A brain-machine interfaceeavesdrops on a small sample of spiking neurons tofirst establish each neuron’s tuning fork (shownhere); and, in subsequent trials, to translate themoment-by-moment pattern of spiking to the monkey’sintended reach direction. Pictures courtesy of:Krishna Shenoy, PhD, Stanford University.where you stick the electrodes. You’re likely to belistening to relevant neural chatter as long asyou’re somewhere in the arm area.EARLY ANIMAL WORK:THE CHANGING BRAINIn 1999, a group of researchers includingMiguel Nicolelis of Duke University, demon-16 BIOMEDICAL COMPUTATION REVIEW Fall

strated a simple BMI system that allowed rats tocontrol a lever with their minds. In 2000,Nicolelis extended this work by having monkeysremotely control a prosthetic arm locatedin a distant city.Since then, several groups have trained monkeysto move cursors or prosthetics in two andthree dimensions using BMI systems. In the labrun by Andrew Schwartz, PhD, professor of neurobiologyand bioengineering at the Universityof Pittsburgh, monkeys can feed themselvesslices of orange and zucchini with a robotic armwired to their brains.The monkeys assimilate the cursors or armswith ease, moving them as fluidly as we would apaintbrush or a tennis racket. “It surprises mehow quickly cells adapt to thinking ‘my purposein life is to move this dot on the screen’ asopposed to ‘my purpose in life is to move thislimb,” Heetderks says.Amazingly, with feedback, the brain learns toadjust the firing pattern of neurons to improveBMI handling. This finding firmly quashes theold theory that brain patterns are fixed in childhood.Brain plasticity will make life easier forBMI researchers, Heetderks says.Some researchers even speculate that theycould scramble the code that transforms neuralspiking to cursor position and, with enoughtraining, the brain could relearn how to fire itsneurons to control the cursor. So far, such testsin monkeys have flopped as monkeys quicklyget frustrated and give up.The plasticity of the brain has surprised scientistsso much that it is no longer ludicrous towonder if neurons from a non-motor area ofthe brain, such as the visual cortex, could drivea prosthetic arm.of bringing brain-interface technology tohumans. The company, Cyberkinetics, Inc., ofFoxborough, Massachusetts, launched its firsthuman trial in 2004 with Matthew Nagle.“We thought that there were certain limitationswith animals that we couldn’t overcome until wegot to humans,” says Cyberkinetics co-founderNicholas Hatsopoulos, PhD, now assistant profes-FROM MONKEYS TO HUMANS:THE LATE STONE AGEIn 2001, four researchers from BrownUniversity formed a company with the missionTELEKINETIC MONKEY Upper panel: During calibration,a monkey uses a joystick to move a cursorand, correspondingly, a remote robotic arm. Allthe while, a computer correlates the pattern ofneural spiking in the monkey’s brain to his joystickmovements. Lower panel: In brain-control mode,the computer directly translates the monkey’sbrain signals into cursor (and robot) movement.Courtesy of Miguel Nicolelis, Duke University.www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 17

“Implantationmethods areprobably inthe late stoneage,” agreesJonathanWolpaw, MD.“Basicallywhat we donow, is weput nails inthe brain.They’re veryfine, tiny,well-machinednails, butthey are infact nails.”BRAIN TAPPING Upper panel: Several motor BMIgroups use a microwire array. Courtesy of MiguelNicolelis, Duke University. Lower panel: Othergroups use the Utah Electrode Array, includingCyberkinetics, Inc. Courtesy of Richard Normann,University of Utah.sor of organismal biologyand anatomy at theUniversity of Chicago.You cannot ask monkeysto do complex cognitivetasks or tell youwhat they are feeling,he explains.Surgeons implanteda tiny array of 100electrodes, each narrowerthan a humanhair, into Nagle’smotor cortex. Theseelectrodes act like 100tiny microphones suspendedin the brain tissue, each able to hear upto a few neurons. Surgery took less than threehours, and there were no complications.To calibrate the system, Nagle is told to movehis hand left or right, as a computer associates thebrain pattern with direction. Once calibrated, thecomputer reads signals out of Nagle’s brain andmoves the cursor accordingly. Soon, Nagle movesthe cursor without thinking about his arm. Hecan check a mock email program, play a simplevideo game, even draw a crude circle. And he canwhistle or talk—tasks that fire other neurons inhis brain—without loosing control of the cursor.“The patient was thrilled initially,”Hatsopoulos says. “Here’s a case where a patientwho hasn’t moved anything in several years canactually move something intentionally.” ButNagle soon wanted more, Hatsopoulos admits.“I can’t say it’s perfect.”Nagle can only use the system when techniciansare available. The system has to be calibratedfor each use, due to implant shifting;also, Nagle’s head has to be plugged into thecomputer via a bundle of wires. Cyberkinetics isworking on automating the system and makingit wireless. Besides Nagle, a second patient hasbeen implanted with the system, but no data areavailable about him yet. Heetderks calls thetrial a “real milestone.” It shows for the firsttime that the system works in somebody who isparalyzed, not just in intact monkeys, he says.But not everyone agrees that the Cyberkineticstrial was a milestone. “We are appalled by what’sgoing on,” says Miguel Nicolelis, MD, PhD, ofDuke University, who did much of the pioneeringwork on BMIs in animals. Nicolelis is professor ofneurobiology, biomedical engineering, and psychologicaland brain sciences at Duke and codirectorof Dukes Center for Neuroengineering.He calls the trial a marketing stunt. Because Naglecould work a computer without the BMI (mechanismsexist for moving a cursor using one's voice oreyes), the benefits of the implant do not outweighthe risks of neurosurgery, he says. Opening theskull and inserting electrodes into the brain is theonly way to record neural spiking, but it risksinfection or bleeding.Nicolelis contends that the trial was also scientificallyunnecessary, since imaging studiesalready showed that quadriplegic patients havea working motor cortex. “We didn’t need to putelectrodes in their heads to answer that.”Nicolelis has his sights set on human trials inthe next one to three years, but he is waiting for18 BIOMEDICAL COMPUTATION REVIEW Fall

a few key pieces: he wants a completely wirelesssystem that can drive a prosthetic arm with severaldegrees of freedom. He says no system outthere is ready for prime time.“Implantation methods are probably in thelate stone age,” agrees Jonathan Wolpaw, MD, aresearch physician at the Wadsworth Center atthe New York State Department of Health, inAlbany. “Basically what we do now, is we putnails in the brain. They’re very fine, tiny, wellmachinednails, but they are in fact nails.”MAKING WAVESNagle is not the first human to control a cursorwith his thoughts, but previous systems hadmuch lower resolution. In 1998, a stroke victimnamed Johnny Ray, who could communicateonly by blinking, received a two-electrodeimplant that allowed him to select letters andicons with his brain. The device read from onlya few neurons, unlike the array in Nagle’s brain.Additionally, since 1991, many people havedriven cursors with their brainwaves, which arerecorded non-invasively. The non-invasiveapproach is more commonly called a braincomputerinterface (BCI).The firing of millions of neurons generates anelectrical noise that can be recorded by electrodesplaced on the scalp. These collective signalsdo not give precise information about threedimensionalarm trajectories, but can be used tocontrol a computer cursor. The impetus of thistechnology is helping people who are “lockedin,”explains Wolpaw, who heads the BCI programat the Wadsworth Center. Locked-inpatients have normal brain function but no musclecontrol—giving them no means of communicatingwith the outside world. The condition candevelop as a result of a brainstem stroke, a highspinal cord injury, or amyotrophic lateral sclerosis(ALS or Lou Gehrig’s disease).BCI systems unlock thesepatients, allowing them to move acursor, surf the internet, and selectletters at a slow rate—up to 1-2words per minute. “If they can doeven a couple of letters a minute,that’s a big step because they’vehad some control returned tothem,” Wolpaw says.The Wadsworth BCI is an electrodecap that fits over the skulland is wired to a computer.Patients begin training by imagin-USE YOUR BRAINWAVESLeft panel: A volunteer usesthe Wadsworth brain-computerinterface to voluntarily move acursor with his brainwaves.Right panel: Using theWadsworth BCI, four volunteersmoved a cursor to 8 targets (2Dcontrol), with varying degreesof accuracy. Colors show cursorspeed: red is fastest; blue isslowest. Courtesy of JonathanWolpaw, Wadsworth Center,New York State Departmentof Fall 2005 BIOMEDICAL COMPUTATION REVIEW 19

“If there isgoing to be abreakthroughin that area,it’s goingto be the‘walking andchewing gumat the sametime’ issue,” movement—running, playing baseball, stickingout their tongue—to depress idling rhythmsin the brain (known as mu and beta rhythms). Acomputer moves a cursor up or down accordingto the strength of these rhythms. After hundredsof trials, most people can voluntarily move thecursor up and down or right and left.Recently, however, Wolpaw’s group hasachieved two-dimensional cursor movement.Using independent control signals from muand beta rhythms, patients can steer a cursor toone of eight target points on a screen, with upto 90% accuracy.The brain is now achieving its intentthrough non-muscular output, Wolpaw says.But control requires concentration, making ithard to multitask.It’s a problem that Heetderks says is a majordrawback of non-invasive systems. “If there isgoing to be a breakthrough in that area, it’s goingto be the ‘walking and chewing gum at the sametime’ issue,” he says.Despite this limitation, Wolpaw’s recent successeshave surprised invasive BMI researchers.“I would claim that they’ve been able to do betterwith their non-invasive electrodes than thevast majority of the invasive electrode work,”says Krishna Shenoy, PhD, who works on invasiveBMIs as an assistant professor of electricalengineering at Stanford University. “It’s harderto motivate monkeys,” he adds.Invasive technology gives quicker learning andmore resolution than non-invasive systems, butthese advantages have not been reflected in significantlybetter performance, Shenoy says. He callsit a “kick in the butt” for the invasive communityto optimize their systems.MOTORING ONElectrodes have a limited shelf life in thebrain; they get gummed up with immunologicalfactors and stop picking up signals after anywherefrom a month to a few years. A big challengefor invasive BMIs is to make implants reliablylast a decade, the benchmark for otherimplantables like cardiac stents. “It’s a bit riskyfor the whole field to be betting on it, but thereare good people working on that,” Shenoy says.Researchers will also need to fit all the electronicsfor sensing, amplifying, and decodingsignals on one tiny implantable chip that canwirelessly transmit to a prosthetic arm or otherdevice. If you have all this stuff in the brain,you’d want a pretty dexterous prosthetic arm,too, Shenoy says, but it does not exist yet.Though BMI research has focused on armmovement, the Holy Grail of motor BMIswould be locomotion, Nicolelis says. Eventuallyhe hopes BMIs will drive an exoskeleton suit torestore walking in someone who is paralyzed.For now, he is working on taking brain signalsfrom rats on a treadmill to drive a robot witharms and legs. To do this successfully, Nicolelismust pull many different control signals frommultiple arrays placed in different regions of arat’s brain. Because he’s reading from multipleareas, he’s also been able to show something surprisingabout what happens in the brain when itcommands an arm to move.Current textbooks present a hierarchical pictureof arm reaching; in short, the posterior parietalcortex (deep in the brain) launches theintention to move; then the pre-motor cortexdraws up an exact plan for moving; then themotor cortex moves the muscles. However, readingfrom these three brain regions simultaneously,Nicolelis has found a different picture: thespiking sometimes appears directly in the motorcortex or simultaneously in all three areas. Touchfollows a similarly surprising path, he’s found.“My students were dying to see a beautifullinear sequential activation pattern—they wanteda Science paper. I said: ‘Sorry kids, we don’tsee it.’ And we got a Science paper anyway.”CONVERSATIONSWITH THE BRAINThe implant in Nagle’s brain was originallydeveloped at the University of Utah for a differentpurpose: to send signals into the brain forartificial vision. Talking to the brain is not muchdifferent from listening to it—you just have toknow the language, the pattern of electricalspiking that causes the brain to hear and see.More than 50,000 people have already had acochlear implant placed in their ears. This devicetransforms sound waves into electric pulses thatthe brain can understand. Current implants have22 electrodes that directly stimulate the neuronsof the auditory nerve in the inner ear. That’senough for some people who were once profoundlydeaf to talk on the telephone.Vision researchers hope to someday matchthis success, but the eye is more complex thanthe ear. To see the letter E, light reflected off theletter is focused onto 100 million photoreceptor20 BIOMEDICAL COMPUTATION REVIEW Fall

Depiction of what you might see with a sub-retinalartificial vision system implanted (black &white box). The black dot in the center representsthe blind spot where the optic nerve enters theeye. Such systems currently provide 10 degrees ofvision in the visual field. Courtesy of DanielPalanker, Stanford Universitycells (cones and rods) at the back of your eye,which convert light into a pattern of electricpulses. Adjacent cells in your eye then compressthe image 100-fold, process it and route it downthe optic nerve to your brain. When the messagereaches the visual cortex, you see ‘E’.Vision BMI systems bypass breaks in thispathway by feeding digitized pictures from aneyeglass-mounted video camera to an implant.Implants in the retina bypass degenerated photoreceptors;implants in the visual cortex bypassdamage to the optic nerve. These systems willprobably not work in those blind since birth,since their brains never learned how to see.Typical implants transmit a pattern of electricpulses that match the pixels of light and dark fromthe incoming image. More electrodes give moreresolution. The brain recognizes this patternbecause photoreceptors in the retina and neuronsin the visual cortex are organized to correspond tothe lay of the land. For example, the top right ofyour visual field is “seen” by a cluster of neuronsin a particular location in your visual cortex.Since as early as 1978, scientists at theDobelle Institute of Long Island, New York, haveplaced electrodes in the visual cortex of a handfulof blind patients. Patients see phosphenes—bright spots of light that scientists deem the “starrynight effect”—that confer enough informationto avoid bumping into objects.Second Sight, Inc. of Sylmar, California, hasrecently implanted a 16-electrode array on thesurface of the retina (epi-retinal) that directlystimulates the optic nerve. Patients can recognizepatterns, horizontal and vertical lines, anddirection of motion, says Daniel Palanker, PhD,who works on artificial vision as an assistant professorof ophthalmology at Stanford University.“It’s a big step forward,” he says. Palanker hasplans for a sub-retinal implant, to directly replacedamaged photoreceptors; this designenters the vision pathway at its earliestpoint, without missing the eye’s naturalimage processing. Such an implant couldconfer 20/80 visual acuity, he estimates,but it is still early in development.To truly recreate vision, the brain needs toreceive multiple images per second. Similarly, tocontrol a prosthetic limb, the brain has to sendout multiple control signals per second. In bothcases, the challenge is to interpret huge amountsof streaming data that are noisy, high-dimensionaland changing over time, says Michael Black,PhD, professor of computer science at BrownUniversity; he teaches a computationally orientedclass on BMIs at Brown.“You have to sit down and try to figure out:how does the activity of these hundreds of cellsrelate to these tens or hundreds of parameters ofbehavior or motion. That requires machine learningand computer science techniques: visualization,modeling, a lot of probability and statistics.”The complexity of the human brain presentsa great opportunity for computer scientists, hesays. “Everyone’s running to get involved incomputational molecular biology at themoment, but the brain is really a growth area forcomputer science,” he says.BRAVE NEW WORLDUnlocking the brain—the very seat ofhuman consciousness—is both fascinating andfrightening. Our imaginations quickly slip thebounds of science and drift into science fiction,conjuring up brain chips that boost memory andintelligence, drive cars and planes, and leave ussusceptible to mind-control.The reality is that brain interfaces are verymuch in their technological infancy, and thefocus is medical. Scientists are working onimplantable chips to enhance memory, but thetechnology—which is still being tested incells—is aimed at those with Alzheimer’s diseaseor head injury. The Defense AdvancedResearch Projects Agency (DARPA) heavilyfunds BMI research with a view toward militaryuses, but the agency isn’t chasing cyborgsoldiers so much as prosthetic limbs foramputees coming back from war.“It’s important to keep that perspective inmind in terms of what we’re trying to do here,”Heetderks says. “It’s to improve quality of life,not to make some kind of a Super Human.” ■To trulyrecreate visionor to controla prostheticlimb, thechallenge isto interprethuge amountsof streamingdata thatare noisy, highdimensionaland changingover time,says Black.www.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 21

editorÕsEditorÕsPickspicksBY DAVID PAIK, PhD, EXECUTIVE EDITORDo you have a few favorite books that you recommend to anyone with aninterest in biomedical computing? Are there software products or web sites thatyou love to evangelize? We’d like to open this forum up to the community formini-reviews of biomedical computation—books/papers/Web sites that havecaught your interest. Email us at SoftwareESR’S CATHEDRAL OR BAZAARThe paper’s cryptic title, The Cathedral and theBazaar, dares the reader to delve in and readfurther. Eric S. Raymond (a.k.a. ESR) is a longtimeadvocate for open source software and theauthor of this classic essay. At its most literal level,it documents the history of email software, but thisis only the framework for the true topic: a firsthandexperiment with the open-source developmentprocess that has made Linux so successful.Raymond describes the traditional closedsourcedevelopment process as a cathedral, witha centralized, selective, authoritative hierarchythat releases ordained versions of software onlywhen ready. On the other hand, he describesopen-source development as a bazaar, with taskspromiscuously delegated among developers withdifferent agendas and approaches.He expands on the topic by enumerating 19 lessonslearned from his experience with fetchmail,about half of which can be generalized beyond programmingand software design. Most interesting areRaymond’s insights intoDETAILSEric S. RaymondThe Cathedral andthe Bazaar culture of the opensourcemovement. He describeshow motivation, teamwork andcommunication differ from thatseen in the “cathedral” setting.Since open-source developmentis a growing trend in biomedicalcomputing, many ofthese observations and lessonsapply directly. But there areother ways in which this essay may also be fitting. Theopen-source movement shares numerous similaritieswith the culture of big science projects, many of whichinvolve biomedical computing in one form or another.Research teams of the future, as outlined in the NIHRoadmap, stress interdisciplinary and collaborativeeffort that echoes many characteristics of the bazaarmodel. Many of the 19 lessons could apply to big scienceprojects as well.One of Raymond’s 19 lessons explains why opensource software improves so rapidly. Dubbed “Linus’slaw” (after Linus Torvalds, who launched the Linuxmovement), the lesson states: “Given enough eyeballs,all bugs are shallow.” In essence, given enough co-developersand beta-testers, a problem (or “bug”) will be characterizedand fixed by someone. It’s an embodiment ofthe Delphi effect, which states that averaged expertopinion is more reliable than a single individual. If thesame lesson proves true for the biomedical computationinfrastructure currently under development, that will begood news for the field as a whole.“If you push mass andvelocity high enough youget surprises like nuclearexplosions or Linux.”—Eric S. Raymond22 BIOMEDICAL COMPUTATION REVIEW Fall 2005

Software EngineeringFINDING PATTERNSBy relying onreference bookssuch as DesignPatterns and itsprogeny, designerscan spend their timecoming up with newideas ratherrepeating others’mistakes.www.biomedicalcomputationreview.orgDon’t reinvent the wheel—soundadvice whatever your profession.Novelists have reused archetypalplots and characters throughout history;graphic artists reuse design templates; and architectsreuse architectural motifs. The same is true for programmers:certain design problems recur, and there’s noneed for every programmer to tackle them all over again.Examining reuse at a higher level of abstraction, this book ismore about idea reuse than code reuse. This book is particularlyrelevant to thiscommunity because itpromotes the use ofsolid design conceptsthat may prove valuablefor building a biomedicalcomputinginfrastructure.Instead of discussingprogramming languagesyntax or features, thisclassic text presentsdesign patterns that arecommonly used motifsin object-oriented programming.After ashort introduction, thebook catalogs 23 of themost commonly usedDETAILSErich GammaRichard HelmRalph JohnsonJohn VlissidesDesign Patterns:Elements of ReusableObject-OrientedSoftware,Addison-WesleyProfessional, patterns,although it does notclaim to be an exhaustivelist. The patterns are then divided on two axes. The purposeaxis distinguishes between creational (dealing withobject creation), structural (dealing with composition ofclasses/objects), and behavioral (dealing with class/objectinteraction and distribution of responsibility). The scopeaxis distinguishes between patterns that deal with classesand objects. Some of the patterns are obvious enough thatmany designers use them withoutneeding to use this book,whereas other patterns may beless commonly understood.Design Patterns has become apowerful and important bookbecause it starts to map out theuniverse of patterns in a singlereference collection. Since 1995,when this book came out, othershave compiled lists of anti-patterns:programming motifs thatshould be avoided. Sometimes,studying poor form can be just asinstructive as studying good form.By relying on reference books such asDesign Patterns and its progeny, designers canspend their time coming up with new ideasrather than reinventing the wheel and repeatingothers’ mistakes.Development EnvironmentFIXING WINDOWS (WITH CYGWIN)Every once in a while, something so ridiculouslyuseful comes along that you feel theurge to go out and tell everyone you knowabout it. For me, that something is Cygwin, a UNIXlikeenvironment that runs within MicrosoftWindows. Mac OS X and Linux may be compellingalternatives, but for diehard Windows users, Cygwincan’t be beat for ease of use.Cygwin is free software and the installer is simple, availableat It gives Windows a commandline shell that allows the user to use most of the standardUNIX utilities you would expect. Because it is not an emulationlayer, programs running within Cygwin run just asfast as native Windows code. Some interesting alternativesexist, such as VMware, coLinux and Bochs, but none ofthese have the native integrationand ease of use of Cygwin.Installing is as simple as runningthe setup program fromthe Web site and selecting theindividual packages that youwant to install; the defaultselection is rather minimalist.Cygwin brings tremendousnew functionality to Windows, including a full X Windowsport that allows X11 programs to run alongside Windowsprograms seamlessly. One can run various servers includethe Apache web server, FTP, SSH/SCP, NFS, and mail.Programming tools include gcc/gdb, perl, python, emacs,vi, make, cmake, doxygen, OpenGL, LAPACK, cvs, subversionand swig. Other miscellaneous applicationsinclude PostgreSQL, gnuplot, octave, grep, tar, cron, wget,ghostscript and TeX.There are, however, a fewlimitations and caveats.Windows and UNIX handlenewlines and carriage returnsdifferently and so care must beCygwin bringstremendousnew functionalityto Windows.DETAILSCygwinhttp://www.cygwin.comtaken in I/O programming. Also note that the Cygwin APIlibrary is covered by GNU General Public License (GPL),which means that if you publicly release programs compiledunder Cygwin, you must make source code available.Cygwin has fundamentally changed my computinghabits, providing a pretty good best-of-both-worlds compromise.Nothing is perfect but this does make significantimprovements to Windows. ■Fall 2005 BIOMEDICAL COMPUTATION REVIEW 23

under the hoodUnder TheHoodBY RAY SOMORJAI, PhDTwin Curses PlagueBiomedical Data AnalysisNoninvasive experimental techniques,such as magnetic resonance (MR),infrared, Raman and fluorescencespectroscopy, and more recently, mass spectroscopy(proteomics) and microarrays(genomics) have helped us better understand,diagnose and treat disease. These methodscreate huge number of features, on the orderof 1,000-10,000, resulting in Bellman’s curseof dimensionality: too many features (i.e.,dimensions). However, clinical reality frequentlylimits thenumber of availablesamples to the orderof 10-100. This leadsto the curse ofdataset sparsity: toofew samples. Thus,on the one hand, wehave a wealth ofinformation availablefor data analysis;on the other hand,statistically meaningfulanalysis is hampered by sample scarcity.Robust, reliable data classification (e.g., distinguishingbetween diseased and healthy conditions)requires a sample-to-feature ratio onthe order of 5-10,DETAILSRay Somorjai is Head ofBiomedical Informatics at theInstitute for Biodiagnostics,National Research CouncilCanada. The three majorthrusts in his Group are supervisedclassification, with specialemphasis on handling the peculiaritiesof biomedical data,unsupervised classification (e.g.,EvIdent, a powerful fuzzy clusteringsoftware) and the mathematicalmodeling of thespread of infectious diseases(AIDS, SARS, etc.).instead of the initial1/10-1/1000. Whatcan be done?To lift the curse ofdimensionality andreduce the number offeatures to a manageablesize, we use featureextraction/selection(FES). FESreduces dimensionalityby identifying andeliminating redundantor irrelevant information.For microarraydata this is accomplishedby first identifyinggroups of correlatedgenes and defininggroup averages asnew features. Forspectra, neighboringfeatures arestrongly correlated,and therefore the majorityof features are redundant. In addition, manyfeatures are “noise,” or are irrelevant for thedesired classification. Eliminating these yields amuch lower-dimensional feature space that sufficesfor accurate spectral characterization. Toidentify the spectral features to be eliminated,Dataset sparsity has more subtleconsequences, and lifting this curse ismore problematic. The ideal solution—acquiring more samples—is frequentlytoo expensive or even impracticable.we have developed an algorithm that helpsselect optimal sub-regions that are most relevantfor an accurate classification. Averagingadjacent spectral intensities leads to furtherreduction, while retaining spectral identity,which is important for interpretability of theresulting features (e.g., MR peaks, essentiallyaverages of adjacent spectral intensities, aremanifestations of the presence of specific chemicalcompounds).Dataset sparsity has more subtle consequences,and lifting this curse is more problematic.The ideal solution—acquiring more samples—isfrequently too expensive or evenimpracticable. Yet, limited sample size may createclassifiers that give overoptimistic accuracies,even after feature space reduction. Robustclassifier creation requires enough samples tomeaningfully partition the data into training,validation and independent test sets. The trainingset is used for both FES and optimal classifierdevelopment. The validation set helps preventthe classifier from adapting to the peculiar-continued on page 2524 BIOMEDICAL COMPUTATION REVIEW Fall 2005

ities of a finite training set (overfitting) by monitoringthe progress of the FES/classifier. Theindependent test set is used for external crossvalidation,but only after completion of the FESand identification of the final classifier. Withsmall datasets, even partitioning into trainingand test set is statistically suspect, and k-foldcross-validation is used: the dataset is split into kequal parts (~5-10), trained on k-1 parts andtested on the remaining portion. One thencycles through k times and averages the testresults. For small sample sizes, the variance ofthe averaged test accuracies tends to be unacceptablylarge, while overtraining is still a threat.For highly imbalanced classes (e.g., rare diseasevs. healthy), overall classification accuracycan be misleading. For example, consider 90samples in the healthy class, but only 10 in thedisease class. Misclassifying all 10 still gives 90%overall accuracy. Hence, balanced sensitivityand specificity (i.e., comparable accuracies forboth classes) is more appropriate, and can beachieved by undersampling, oversampling or bypenalizing misclassifications differently for differentclasses. (Differing misclassification costsfor the classes is an example.)For each sample, we compute class probabilities.This is relevant clinically (e.g., additionaltests would be suggested if a classifier assigned apatient to the disease class with 55% probability,immediate treatment would commence ifthis probability were 90%.)In the biomedical field, the twin curses aregenerally active. They both must be dealt within concert, otherwise overly optimistic and frequentlywrong conclusions will result. ■SeeingScience continuedEcce Homology is aphysically interactivenew-media work thatvisualizes genetic dataas calligraphic forms.With a name inspired by FriedrichNietzsche’s Ecce Homo, a meditation onhow one becomes what one is, the projectexplores human evolution by examiningsimilarities between genes fromhuman beings and a target organism, inthis case the rice plant. Ecce Homology isa physically interactive new-media workthat visualizes genetic data as calligraphicforms. A novel computer-vision basedinterface allows multiple participants,through their movement in the installationspace, to select genes from thehuman genome for visualization usingthe Basic Local Alignment Search Tool(BLAST). Five projectors present thesechanges in Ecce Homology’s calligraphicforms across a 40-foot wide wall.“If we worked on the genomic calligraphyvisualization further, it couldbe useful to scientists,” she says, “butthe installation is not a tool; it’s art.And it’s specifically ambiguous and abit mysterious—by intention.”Ecce Homology, which was first displayedtwo years ago at the FowlerMuseum in Los Angeles, works on manylevels both scientifically and artistically.“People assume that there’s value in thevast amounts ofgenomic data we aregenerating,” says West,“but data is not knowledge,and in order forus to derive knowledgefrom it, we need tointerpret it. The morecomplex it is, the harderit is for humanbeings to do that and,consequently, thegreater our need to findnew approaches.” So,says West, “we’ve producedan artwork thatboth speaks to thisneed and lets viewersinteract fluidly with thedata in a visceral way.”Ultimately, West says, the exhibitposes the question, “If you were to dowork that’s truly hybrid art/science,what would that process be like? Andwould there be any outcome that wouldpoint to how art might nurture scientificdiscovery?”For more information about EcceHomology, visit ■Ecce Homology’s custom softwaretransforms strings of genetic codeinto luminous, scientifically accuratevisualizations that incorporatemultiple biological features. Forprotein sequences, the strokeplacement, shape and brush qualityare determined by physical andchemical properties, such as theproportion of mass to volume,hydrophobicity, or ionization of theamino acids. The visualization iscreated from amino-acid sequencechunks that are segmented by a“turn prediction” algorithm. Eachsegment’s corresponding calligraphicstroke is connected to itsneighbor by a connection whoseshape is based on a secondarystructure property of the segment.The result resembles calligraphy.Courtesy: Ruth Westwww.biomedicalcomputationreview.orgFall 2005 BIOMEDICAL COMPUTATION REVIEW 25

D I V E R S E D I S C I P L I N E S , O N E C O M M U N I T YREVIEWNonprofit Org.U.S. Postage PaidPermit No. 28Palo Alto, CABiomedical Computation ReviewStanford University318 Campus DriveClark Center Room S170Stanford, CA 94305-5444seeing scienceSeeingScienceBY KATHARINE MILLER, MANAGING EDITORArt ThatÕs A BLASTAgroup of artists and scientistshas created an interactive artworkusing BLAST (BasicLocal Alignment Search Tool), one ofthe foundational algorithms for comparativegenomics. Normally, the BLASTprocess of looking for homologous genesis invisible to researchers: They enter agene sequences into a computer andwait for BLAST to kick out the matches.But the collaborative art work opensthat black box by making the BLASTsearch visible. The result, EcceHomology was shown in early August atSIGGRAPH 2005 in Los Angeles.“It’s artwork that’s deeply grounded inscience without being didactic,” saysRuth West, director of visual analyticsand interactive technologies at theUniversity of California, San Diego,National Center for Microscopy andImaging Research and research associatewith the UCSD Center for Research inComputing and the Arts. She heads thecollaboration which includes 11 biologists,artists and computer scientists fromUCSD, UCLA and the University ofSouthern California.continued on page 25Visitors interacting with theEcce Homology installationat the Fowler Museum.

More magazines by this user
Similar magazines