Psychiatric Genomics, a drug discovery start-up located in Gaithersburg, Md, will be employing high-throughput screening of tissue samples from psychiatric patients to develop a proprietary gene expression database in which normal, diseased, and drug-associated patterns in mental illness can be compared.
The company grew out of an incubator assembled earlier this year by Alan Walton, a general partner at venture capital firm Oxford BioScience Partners. “We wanted to determine whether it was time to start solving the disorders associated with psychiatric illness. We came to the conclusion that it was indeed time,” said Walton, who as chairman of psychiatric genomics helped to raise $12 million to begin operations. The company is now putting technology in place with the aim of clarifying the role of multiple genes in diseases such as bipolar disorder, autism, and schizophrenia.
Since polygenic disorders pose a challenge to current bioinformatics tools, Psychiatric Genomics CEO Michael Palfreyman said the company’s technology platform would be built specifically to address this problem. Licensing or access arrangements will eventually be available to pharmaceutical companies and other potential partners.
Psychiatric Genomics is planning to build its proprietary gene expression database using brain and blood samples from psychiatric patients, obtained from tissue repositories under exclusive agreements. The company has also subscribed to a central nervous system-specific segment of Gene Logic’s GeneExpress database, called GeneExpress DataSuite.
“From expression analysis, we can tell which genes turn on and off” under various conditions, Walton said.
Les Klimczak, director of bioinformatics and database technologies at Psychiatric Genomics, said that he has taken a “data-centric” view. “The major challenge isn’t to come up with a fancy tool, it’s to get the right data in place,” said Klimczak.
Klimczak said the fragmentation of biological data is a significant problem in data mining efforts. “The databases don’t talk to each other,” he said. “You need integration tools to use them to the full extent.”
He said Psychiatric Genomics’ strategy would entail designing experiments a priori so that biochemical, gene expression, and cell model data flow into a structure that can be mined from a single point. For example, scientists will be able to cross-reference patterns of gene expression in tissue from patients with mental disorders and similar patterns observed in cell lines exposed to certain chemicals.
Klimczak’s goal is to develop the right data model, define it in a sufficiently universal way for it to work with a variety of programming languages or data management systems, and help Psychiatric Genomics scientists build it into their experimental design.
So what’s the right data model? One that describes biological reality, Klimczak said. It needs to incorporate clinical variables, so that parameters that may vary widely in the samples don’t simply appear as noise. When working with samples from tissue repositories, therefore, the degree and consistency of annotation is important.
“We want a detailed description of a patient’s condition, not just ‘sick’ or ‘well,’” he said. Sufficient clinical data will allow researchers to look at a patient’s environment, family background, and medical history in combination with changes in gene expression, so that it can be determined which groups of factors the genes correlate to.
In addition, the model will be designed in such a way that it will be convenient for researchers to extract data and to use data mining tools. “We’re not just building a data store,” Klimczak said. A layered system in which complex levels of metadata are needed to describe the physical location or other “mechanistic” attributes of the data is not suitable, he said.
Instead, the metadata should incorporate descriptors or categories that provide an understanding of the underlying data. Typical laboratory information management systems don’t handle this very well, Klimczak noted.
Psychiatric Genomics will probably employ a combination of commercial and custom software, Klimczak said. Commercial tools such as Partek’s pattern recognition software, which Klimczak praised for its built-in statistical procedures and wizards, may be useful. He also expects to develop simulators and other domain-specific software from scratch. In addition, the software “glue” to integrate the various software modules and fit legacy data into Psychiatric Genomics’ data view will be written in-house.
For mining large quantities of data, he sees the need for better computational tools. Software development efforts have stressed user-friendly “point-and-click” packages. But user-friendly isn’t necessarily computer-friendly. With increasing numbers of projects generating copious amounts of data to crunch, SQL-based systems or other robust tools are needed to facilitate batch processing, using programming interfaces, and specific attributes rather than keywords, he said.
Better computational back ends are also needed to support visualization tools and other end-user interfaces. “When you have large and heterogeneous databases, it’s hard to come up with visualizations that will be helpful in one step. With 60,000 genes, you lose resolution in visualization. Computational strategies at least narrow down the list,” he said.
Klimczak envisions an iterative software development and systems integration process, with input from company scientists at each step. Researchers will have the opportunity to critique the systems under development, as well as present their own needs and help to brainstorm solutions. “We’re a small company,” Klimczak said, “and we’re starting to build in an optimal way.”
Derek Hook, Psychiatric Genomics senior director of genome drug discovery, is in charge of putting the high-throughput screening and robotics systems into place. In contrast to most current systems in which a single measurement is taken per microplate well, Hook said, Psychiatric Genomics is interested in the new high-content multiparametric screening systems, such as those manufactured by Cellomics.
Cellomics’ system is based on high-resolution fluorescence imaging of multiple targets in intact cells. However, Hook said he has not yet settled on a particular hardware platform, and he expects to be integrating equipment from more than one manufacturer.
—Sherri Chasin Calvo