As genomics vendors struggle to pry open pharma’s purse strings, GT gets seven senior pharma managers to tell you just what’s on their wish list
By John S. MacNeil
There’s no doubt that pharma needs new technology that can speed up drug discovery and development and reverse the trend toward ballooning R&D costs. So far, genomics has provided a return on investment, turning up more drug targets. But pharmas are looking for more, and genomics technology vendors are knocking themselves out trying to come up with the miracle tool to win pharma customers. If you are among those dying to know what will make big pharmaceutical buyers happy, read on.
Pharma bigwigs might not be the most candid people when it comes to talking about future plans, but this month GT cornered seven of them to ask about their most mundane wishes and wildest dreams for new technology in genomics, proteomics, and bioinformatics. The result, as you’ll see in the following pages, is a mishmash of ideas about which new tools might help drug discovery — from more powerful protein identification software to a system for organizing genomic data using standardized parameters.
Vendors might complain that big pharma isn’t buying new technology the way it used to, but so far there are few outward signs that eroding profits have bitten significantly into R&D budgets. According to Stephens, an investment bank in Little Rock, Ark., average R&D spending for eight of the largest US pharmaceutical companies grew 6.8 percent during the third quarter, mirroring past trends that show spending growth over the previous year accelerating as the calendar year progresses. This rosy picture could easily fade, but the honchos GT interviewed express confidence that their ability to acquire new technology will continue unabated.
The responses GT collected represent more than just wishful thinking. With many genomics and proteomics companies struggling to turn a profit, it certainly pays to know what pharma’s R&D managers are looking to buy.
Jim Fickett, Global Director of Bioinformatics, AstraZeneca
Would pay for data on protein pathways and associations with disease.
Wanted: A Decent Protein Pathway Editor
Despite alliances with Gene Logic, Genaissance, Silicon Genetics, and Accelrys, Jim Fickett wishes there were more his bioinformatics tools could do. More and more of the information his colleagues need is not in highly structured databases like GenBank or SwissProt, but out in the scientific literature, “and it’s difficult to find,” says Fickett, AstraZeneca’s global director of bioinformatics.
“On my dream wish list, my wish-that-I-were-in-heaven wish list, is a table of how this gene or this protein is related to [another] protein, what’s the evidence underlying [such a] pathway database, and how this protein plays a role in a disease at the level of molecular pathophysiology,” he says.
Fickett, a GenBank co-founder who joined AstraZeneca in October 2001 from GlaxoSmithKline, also wants a “decent editor” for visualizing the interactions of proteins in pathways and linking the picture with its underlying data. “You can’t easily build a picture where you can click on a link between two proteins and find out which papers present evidence to support that interaction,” he says.
He’d also like a product that can track connections between proteins and diseases. This dream tool would identify reports in the literature that draw on techniques such as immunohistochemistry and knockout models to show a potential causative relationship between a protein and a disease. “That [information] might either increase or decrease our interest in that protein target,” Fickett says. “It would be nice to have that summarized and delivered to us.” He points to Ingenuity, of Mountain View, Calif., as one company interested in working in this area.
As for acquiring these capabilities, Fickett would rather buy than develop software in-house. “We’re very pragmatic,” he says. “We try to estimate what it would cost us to do ourselves, and if we can buy for less we’ll buy it, keeping in mind that usually your estimate to build it yourself is low.”
Dan Davison, Associate Director of Bioinformatics, Bristol-Myers Squibb
Wants a solution to the naming problem.
One Name is All You Need
To Dan Davison, associate director of bioinformatics at Bristol-Myers Squibb, there is one problem that confounds all his other attempts to develop bioinformatics technology in support of BMS’s drug discovery operation: naming. If he had just one wish, he says, it would be to once and for all find a universally accepted name for every gene, splice variant, protein, and post-translationally modified protein.
Especially for pharmaceutical researchers wary of pouring money into the study of targets around which their competitors may already have built programs, the need is clear. “It doesn’t matter if you can study one protein a day or one million if you don’t know what you’ve got,” Davison says. “These are difficult and essential issues, beyond which everything else becomes finding the right screwdriver.”
The name game is not a puzzle for BMS alone, as Davison is quick to add. He points to the Gene Ontology Consortium, which is “starting to give us a way through on some of these problems,” though he admits that his interest only goes so far as BMS. “We don’t try to fix the problem for the world. We try to fix it for our drug discovery programs,” he says.
That sums up how Davison’s group confronts technology development in general. Taking a supported version of the Ensembl database developed by Biotique Systems, BMS scientists tailored the platform to their own requirements. “[Biotique is] routinely amazed at what we do to their stuff,” Davison says. “That thread runs through everything we do with our external vendors.”
Davison, who joined BMS in 1997 after simultaneous academic appointments at the University of Houston, Baylor College of Medicine, and Texas A&M, says the next challenge in bioinformatics will be to devise software and databases for integrating data from metabolic profiling and systems biology experiments. Davison doesn’t expect to be satisfied with an off-the-shelf solution. Although he says he occasionally finds technology that looks interesting, most software available through vendors “is not for our scale.” His budget, he predicts, will remain stable as his group begins addressing new problems.
Michael Man, Senior Statistician, Pfizer
Says vendors lag behind in technology by a year or two.
A Tough Sell for Bioinformatics Providers
Pfizer’s Michael Man has a simple request of bioinformatics vendors looking to satisfy big pharma researchers: “What might help is if the manufacturers have the scientists’ needs in mind,” says the Ann Arbor, Mich.-based statistician.
Man’s wish list reflects his skepticism about the capabilities of off-the-shelf data-analysis software: instead of listing products, he itemizes ways he can fulfill his desire to develop in-house or collaborative projects with academics. “It’s very difficult to subscribe to a vendor because their technology lags behind by a year or two,” he says.
As a statistician who helps Pfizer researchers design and analyze gene-expression-profiling and real-time-PCR experiments, one of Man’s detested chores is adapting output from vendors’ platforms. He must put data from his Affymetrix GeneChip and Applied Biosystems’ TaqMan real-time PCR platforms through multiple manipulations before he can import the information into his programs for statistical analysis, a process that he says is both “error-prone and time-consuming.”
That’s one reason Man says he has begun to embrace open-source software for analyzing expression-profiling data, primarily in the form of a program called Bioconductor. The software, developed under the stewardship of biostatistics researchers at the Dana-Farber Cancer Institute, consists of a set of algorithms for manipulating and normalizing DNA microarray and other genomic data, generating graphics, and annotating the results, among other capabilities. “The package can analyze raw data from the Affymetrix system, and in the packages there are [all] kinds of pretty neat stuff,” Man says.
He would also like to develop a distributed computing system for performing intensive analysis of gene expression and other genomic data. His engineers are attempting to develop the technology in-house, he says, because they can more easily support and customize open-source software. “If we lock in with an external solution provider, we lose the alignment with our internal IT support,” he says.
Man says he has been satisfied with some vendors’ clustering-analysis software for generating colored charts showing expression profiles, but even that software is “pretty rigid.” Open-source software or his own scripts can perform the same task, he adds, and allow him to add more information. “That’s why I don’t subscribe to one particular vendor,” he says.
Hanno Langen, Head of Proteomics Initiative, Roche
Verifying that Proteomics Is on Track
As former GeneProt CEO Cédric Loiret-Bernal once said, there may be 30 pharmaceutical companies worldwide that could build their own proteomics programs, but “there aren’t 30 Hanno Langens.” Few big pharmas can rival the diversity of Roche’s collection of proteomics technologies. With Langen heading up the company’s proteomics initiative, Roche has acquired skills in subcellular fractionation, separating proteins using both gel electrophoresis and non-gel chromatography methods, and automating MALDI-TOF/TOF mass spectrometry analysis, to name just a few.
Now, Langen says his researchers need better tools for identifying proteins by matching mass spectrometry data with information in protein databases, as well as better software for making accurate estimates of the probability that the match is correct. They’re spending “quite a lot of time manually going through the lists that are generated by this type of software available today. There is definitely a need for automation to get more reliable results in the end,” Langen says.
But willingness to buy doesn’t mean much if the technology doesn’t exist.
Historically, his group has tried to address its software needs predominantly through in-house efforts to maintain access to source code, but Langen would not be averse to buying protein identification software off the shelf if it were up to par.
Another area in need of attention: verifying results from experiments that attempt to link proteins with a specific disease or cell state. A readily available supply of high-quality antibodies is crucial, he says. “There is a clear need to verify some of the proteomics results we are getting, and we need antibody tools to do this,” he says.
Langen’s budget for developing and acquiring new technology has increased by a factor of three or four over the last several years, he says, but he must still justify each outlay. “When we can convince management that the technology we want to acquire is better than another technology somebody else is proposing, then we win,” he says.
Dalia Cohen, Global Head of Functional Genomics, Novartis
Wants help managing and extracting information from increasingly complex data.
Looking Ahead to Systems Biology
There’s no question in Dalia Cohen’s mind that genomics has contributed positively to drug discovery. But there are also quite a few things the global head of functional genomics for Novartis wishes it did better.
Measuring the changes in phenotype that result from treating cells with specific small molecules is an established technique for testing new drugs. But how to determine which genes respond to a drug? That’s a capability many pharma researchers have yet to master, she says. Biochemical assays for studying the response of gene expression to a compound are cumbersome, and proteomics techniques, such as pinpointing changes in post-translational modifications, are still too slow.
“There are methods, but the methods are low-throughput, and not very sophisticated,” Cohen says. “This is something which is really missing in this industry.” While there are promising technologies for acquiring this type of expression data, such as protein or small-molecule arrays, they are early stage.
Cohen says she’s also interested in new “knowledge management” strategies, or ways to tease out biological insight from the massive amount of data her group generates through its RNA profiling, proteomics, and functional assay experiments. “The complexity of these data increases on a daily basis,” she says. “We have a major effort internally, but [this problem affects] the whole world.”
Perhaps the biggest gap between the pharmaceutical industry’s current and necessary capabilities, she adds, is in systems biology. In addition to complex algorithms, creating models of biological processes that span genes, proteins, and cellular function will require input from all types of biologists, as well as mathematicians and statisticians, Cohen says. “[We’ll need] algorithms, lab experiments, and the ability to test those algorithms in silico and on the bench,” she says.
Cohen stresses that Novartis won’t try to build all these capabilities on its own. The pharma typically devotes about 30 percent of its total R&D budget to partnerships. New technology development “should be a shared effort,” she says. “You need a lot of brains and a lot of approaches. This is why we believe here that the best way to do it is by collaborating with academia.”
Klaus Lindpaintner, Head of Genetics and Roche Center for Medical Genomics, Roche
His Holy Grail: technology for “transient modulation” of gene expression in organs.
Warning: Mining Garbage Yields Garbage
As head of the Roche Center for Medical Genomics, Klaus Lindpaintner has had ample opportunity to gauge how well genomics is applied to drug discovery. Though it’s adept at identifying potential drug targets, he says, genomics has yet to provide a sufficiently fast and effective method for suppressing gene function.
“The Holy Grail at this point is a technology to do transient modulation of gene function or gene expression, potentially in an organ-targeted fashion, to perform assessment and validation,” he says.
One problem, he adds, is that knockout methods that suppress gene function during an organism’s development tend to introduce other changes that complicate the experiment. Lack of a suitable gene-modulation technology, Lindpaintner says, has become the rate-limiting step to assessing all the new targets that pharma researchers have identified over the last few years.
In proteomics, Lindpaintner sees the primary bottleneck as researchers’ inability to sufficiently reduce the complexity of samples prior to analysis. Typically, he says, researchers “just mince up whole organs, and clearly that results in a very large signal-to-noise ratio.” Improving the value of proteomics experiments will require significant advances in sample fractionation, he adds.
Lindpaintner would also like to see improvements in technology for modeling protein structures and predicting interactions, but he’s a skeptic when it comes to the value of large-scale data-mining efforts. “There’s this idea that we’re going to create this humongous database out of God knows what sources, without looking right or left, and then eventually by mining this database we’re going to come up with meaningful information,” he says. “That may happen now and then, but experience shows if you have garbage in, you’ll have garbage out.”
Lindpaintner is also circumspect about the role of genomics in drug discovery in general. He considers genomics, while indispensable, just one tool out of many required to make new medicines. For this reason, his share of Roche’s R&D funding for new genomics technology is unlikely to grow significantly, he says. Instead, his group will adopt new techniques on a case-by-case basis.
Mark Cockett, Executive Director of Functional Genomics, Bristol-Myers Squibb
Wants reduced sequencing costs and RNAi reagents for every gene.
Straight to the Cellular Assay
Since 1998, when Bristol-Myers Squibb first began investing heavily in genomics technology, the company’s alliances in tech development have blossomed. BMS has drawn on the collective expertise of its in-house scientists and upward of seven collaborators to build capabilities in RNA expression profiling, knockout mouse models, SNP genotyping, antisense technology, and most recently RNA interference.
Given this smorgasbord of tools to choose from, perhaps it is not surprising that a technology wish list for Mark Cockett, executive director of functional genomics at BMS, is relatively brief. First and foremost, he’d like to see the cost of sequencing drop drastically, preferably to the much-bandied-about $1,000 genome that would allow his group to avoid genotyping in favor of sequencing. His other primary concern is to build a library of reagents for performing RNA interference experiments on every target-class gene.
To generate a candidate list of gene targets, BMS scientists typically perform initial experiments involving subtraction libraries from cells, gene chip experiments, or data mined from the literature, says Cockett, who came to BMS almost three years ago from Wyeth. “But ideally, if we had every target-class gene as a verified RNAi reagent, we could simply [skip the first steps] and just go straight to the cellular assay,” he says. “[By running] the functional assay, you’d get an immediate hit as to [which] genes are really involved with that pathway.”
Although Cockett says his budget for acquiring new genomics technology — either by developing techniques in-house or through outside collaborators — has stabilized after several years of growth, he doesn’t predict any change in BMS’s strategy for balancing the contributions that come from within and outside the company. “In terms of preclinical genomics — target ID through target validation through supporting pipeline activities — if I look at my group’s budget for the year, and I look at what we spend externally, it’s approximately 50-50.”