Biomarkers were on the minds • or at least in the press releases • of a number of vendors at last week's American Society for Mass Spectrometry conference in San Antonio. This emerging area of the proteomics sector is viewed as a key driver for growth in the mass-spec market over the next decade, and several vendors who launched new biomarker-discovery platforms at ASMS told BioInform that bioinformatics will be the key to their success in this competitive environment.
Mass spec-based biomarker-discovery platforms are built upon three key components, according to Mary Lopez, business leader for PerkinElmer's analytical proteomics business: a high-resolution mass spectrometer, good sample enrichment, and statistically rigorous analytical software. "It's like a three-legged stool • you take one piece out, and you fall down," she said. "You need the high-resolution detector, you need the high-throughput and specific sample enrichment, and really you need to have the right analysis or your data will be meaningless because you won't be able to come up with a classification model that allows you to use it as a diagnostic to differentiate disease from normal."
Mark Flocco, business development manager for clinical proteomics and biomarker discovery at Bruker Daltonics, agreed that bioinformatics will be the key to differentiating these competing platforms. "The instrumentation is all getting better, more robust, more responsible," he said. "They're all going to be generating large sets of spectra and peaks and information, and it is the people who can mine that information and turn it into knowledge that are going to succeed faster."
Flocco added that "the evolution of our bioinformatics package • going along with the customers' feedback on what additional things they will need • is going to be imperative for growth in this business."
Bruker and PerkinElmer, along with Applied Biosystems, all released biomarker discovery packages to analyze data from their mass spectrometers at ASMS (See below for additional software products launched at ASMS). Bruker launched version 2.0 of the ClinProTools package that it initially launched in late 2003. The release includes a new genetic algorithm called QuickClassifier that the company claims can speed the model-generation process to generate a classification result in only a few seconds, even for very large data sets.
ABI, meanwhile, launched MarkerView, which the company said is able to profile both proteomic and metabolomic biomarkers. "We've taken the approach that many labs, many pharmaceuticals, have in that they have combined their efforts in protein and small-molecule biomarkers to create a biomarker facility," said Julie Wingate, product manager for ABI's small-molecule business. "So we've taken that same approach in the MarkerView software to make it applicable for protein and metabolic biomarkers."
PerkinElmer looked outside its own walls for its biomarker software offering. The company's BioXpression biomarker platform relies on BAMF (Biomarker Amplification Filter), a software package developed by Predictive Diagnostics, the diagnostics subsidiary of Large Scale Biology. Predictive Diagnostics uses BAMF in-house for its own diagnostic development, but has an exclusive relationship with PerkinElmer to provide biomarker analysis services for PerkinElmer's mass spec customers using the software, which runs on a Linux cluster at the company's Vacaville, Calif., headquarters.
"It's not the type of software that can run on a laptop," said PerkinElmer's Lopez. "It's been designed not for looking at 10 to 20 samples, but for doing real statistically significant projects and to deliver specificity and sensitivity values that a researcher or a clinician can place high confidence in."
Despite the variations in marketing and delivery, all these software packages accomplish essentially the same task: They use statistical algorithms to compare the peak lists generated for multiple samples in order to distinguish the differences between them. However, it's the choice of statistical algorithm • as well as rigorous QA and QC • that can mean all the difference in evaluating the performance of different software packages, however.
Guy della Cioppa, vice president of business development at Predictive Diagnostics, noted that many analytical packages can be thrown off by the inherent bias in a sample set, or by a technical issue with a sample plate or the mass spec itself. "If you have a sub-par informatics discovery platform, you can essentially train and test on the systematic errors that might be in the system, thinking that those are valid biomarkers, and then you've essentially built a house of cards for a diagnostic model • it's fine on the limited data set that you're using to validate, but if it doesn't hold up to blinded patient samples, then you know that you've got a sub-optimal model," he said.
ABI's MarkerView relies on t-tests and principal components analysis for its statistical analysis; Bruker's ClinProTools uses QuickClassifier as well as a support vector machine-based approach; while Predictive Diagnostic's BAMF uses a combination of 20 statistical algorithms, according to PerkinElmer's Lopez.
A Booming Market, but
No Bioinformatics Business?
Market research firm D&MD has estimated that the market for diagnostic products based on proteomic and metabolomic biomarkers will grow from a current level of $100 million to $2 billion by 2008. None of the vendors that BioInform spoke to for this article were willing to pin a number on what segment of that market might belong to discovery software, but they all said they expect biomarker discovery to drive growth of their mass spec platforms over the next few years.
"I don't have the numbers for what percentage of our business now is made up of biomarkers or metabolomics, but it's definitely quite a growth area," said ABI's Wingate.
Bruker's Flocco said that the biomarker market is just beginning to enter a "hockey stick" growth curve, "because we are just scratching the surface" now. "The deeper we can dig, the more we can unpeel the onion, get deeper and deeper into it, this market will continue to grow and you will probably see more diagnostic biomarkers developing as we get a better understanding of the human proteome."
While warning that "biomarker" is "obviously a buzzword and everyone's jumped on it," PerkinElmer's Lopez said that the market is still "growing fast." The company views proteomic and genomic diagnostics "as one of the biggest growth areas for the company," she said.
But despite a promising future for the software that will enable this growth, the field of biomarker discovery informatics has so far been led by mass spec vendors • not bioinformatics software providers.
"There are a few software packages out there that could be adapted to do this type of analysis, but there really haven't been any vendors who have come out with biomarker analysis packages," said Lopez. "I think that's because of the sophistication that's required in the algorithms and also the data-processing power that's required in the platforms, and finally the fact that different mass spec vendors are going to have spectra with varying degrees of resolution." She said that depending on the mass spectrometer, an experiment could generate anywhere from 30,000 data points for some of the low-end instruments, to 800,000 data points for a higher-end platform.
Other Bioinformatics Products
Released at ASMS 2005
|A number of companies released software products at last week's American Society for Mass Spectrometry conference in San Antonio. In addition to the biomarker software discussed above, the following product launches were announced at ASMS:|
|Rosetta Biosoftware launched Rosetta Elucidator, a protein expression data-analysis system that supports mass-spectrometry-based, gel-free protein biomarker discovery. The system includes raw data management, LC/MS data processing for quantitative and differential analysis, protein identification, and high level analysis at the peptide and protein level. The product line "will be available in flexible configurations, designed to meet the needs of both large and small organizations," Rosetta said in a statement.|
|Proteome Software released Scaffold, a proteomics software tool for "extracting more detailed and accurate information" from Sequest or Mascot results. Scaffold is able to cross-validate protein identifications using different search tools by probabilistically combines the results from different platforms, "yielding increased sensitivity and higher confidence in the identified proteins within the sample." An evaluation copy is available at http://www.proteomesoftware.com.|
|Bruker Daltonics released Proteineer-LC 1.1, a software package for LC-MS/MS-based quantitative proteomics, and ProteinScape 1.3, which integrates the Mascot, Phenyx, and Profound protein search engines. Bruker co-developed ProteinScape with Protagen. The software supports the mzData standard from the HUPO Proteomics Standards Initiative.|
|GE Healthcare released DeCyder MS, a software package for analyzing LC-MS-based differential protein expression. DeCyder MS represents LC-MS data as 2D or 3D signal intensity maps, enabling researchers to visualize trends, post-translational modifications, as well as other differences between large data sets. All the acquired MS/MS data remains embedded in the signal intensity maps and can be used for optional protein identification, the company said.|
ABI's Wingate echoed Lopez' comments. "There are not at the moment a lot of software packages available specifically for biomarkers and metabolomics," she said. "There are a lot of pharmaceutical companies who are using some in-house resources to develop software, but only out of necessity, because they haven't had tools already commercially available." Wingate said that some researchers have modified general-purpose statistics packages to analyze their spectra, "but they're not designed specifically for the biomarker or the metabolomics workflows."
At least one bioinformatics vendor has identified biomarker discovery as a promising area, however. Protein informatics firm Nonlinear Dynamics announced in April that it plans to launch a version of its Phoretix software for biomarker analysis later this year as the first in a broad range of biomarker informatics tools it is planning [BioInform 04-18-05].
"Many pharma and biotech companies have reconfigured their research models to incorporate biomarker functionality. However at present, widespread implementation of this technique is hampered by the lack of accurate, hardware-independent software commercially available to researchers," said John Spreadbury, group sales and marketing director of Nonlinear, in a statement at the time.
James Sykes, operational marketing manager for the company, told BioInform that the market opportunity for software in this area is hard to pinpoint. "The 'biomarker-discovery software market' is a little bit of a misleading name, since comparing traces in this way is just one way to discover biomarkers • there are still other proteomic methods, some already existing, such as the traditional 2D/MS workflow, and others no doubt in the pipeline," he wrote in an e-mail. "So, yes, we think this stream will certainly be a fast growth area over the coming few years. However, what will really precipitate significant growth in this market as a whole is the ability to identify and validate biomarkers by using multiple concurrent techniques • what we call cross technique analysis."
One advantage that Nonlinear may have over instrumentation vendors is that its software will work with data from multiple mass spec platforms. ABI, Bruker, and PerkinElmer all said that their biomarker software packages will only run on their own instruments.
The first component of Nonlinear's biomarker suite, a medium-throughput analysis tool aimed at the academic market, is scheduled for launch in the third quarter.
• Bernadette Toner ([email protected])