Information Overload

Modern medicine is generating new information at a dizzying pace,
but Children’s researchers are developing the tools and techniques
to make sense of the flood of data.

By Cyril Manning

Picture this: a vast checkerboard, stretching in every direction to the horizons of your imagination. There are roughly 30,000 spaces in this grid – one for every snippet of chemical information that makes up the human genome. Moment by moment, a scattered, apparently random sample of these spaces begins to glow, each fluorescent beacon burning at a specific intensity. It’s a secret code of sorts, genetic marching orders that tell living cells how to behave.

For years, researchers knew this code existed, but finally it has been cracked, at least in a sense. Today, a scientist can buy a small manufactured chip, called a microarray, containing this grid of genetic pieces. By physically extracting RNA (the chemical messenger that transcribes genetic information) from the nucleus of a cell, dyeing it with a fluorescent compound, and dropping it onto the chip, the scientist can see exactly what signals are being sent. That’s because the fluorescent RNA binds with certain pieces in the grid, lighting them up like microscopic beacons.

Since the mapping of the human genome in the 1990s, the problem for scientists studying genetic disease is no longer seeing these genetic messages, but making sense of them. Recall that immense checkerboard: 30,000 spaces, some of them glowing, each at its own intensity. This is only a snapshot of which genes are being expressed (that is, which ones deliver their encoded messages) at one moment in time in one individual. Now picture thousands of these grids stacked one upon another, snapshots of gene expression over time; then overlay them one on top of the next, one for each individual in a population. There are patterns here; the challenge is deciphering them.

Alvin Kho, PhD, can create pictures of the genes that are active in a human brain cancer and compare those to pictures of gene activity during normal brain development. The result? Genetic messages found in early brain development cluster near the genetic messages seen in tumors. Genetic messages found late in brain development, however, don't appear in tumors. This suggests that similar genetic instructions are involved in both.

“When we look at how these genes might be interacting with each other to cause disease, there are billions of possible combinations,” says Isaac Kohane, MD, PhD, director of the Children’s Hospital Informatics Program (CHIP). “It’s like the ultimate Rubik’s Cube.” But it’s a puzzle Kohane and others within the program, which uses a wide range of tools, computer power and analytic methods to address various issues in medicine, plan to solve. The CHIP researchers are sifting through vast amounts of health-related data, identifying previously indiscernible patterns in complex systems, and turning those patterns into valuable insights – not just in genetics, but also in basic biology, diagnosis and treatment of disease, and management of public health issues.

This is no small ambition. It requires investigators with expertise in two or more very different fields. “Math and computer science do not come naturally to most biologists,” says Kohane, himself an endocrinologist and computer scientist. “Many CHIP researchers are dually or triply trained in medicine and mathematics or computer science, and that puts us at the center of some very exciting work.”

Mathematics, not microscopes
One example is the research of Alvin Kho, PhD, which focuses on using computer analysis to understand the flawed genetic instructions that can lead to pediatric cancer – specifically, the brain tumor known as medulloblastoma. A mathematician by training, Kho analyzes these genetic signals in ways that cannot be matched by a traditional biologist looking through a microscope. As Kohane puts it, “There’s a whole generation of biologists who can’t do state-of-the-art work because the com- putational aspect of it is out of their reach.”

For over 100 years, biologists have speculated that there is a close correlation between human development and the process of cancer growth, known as tumorigenesis. Although lacking proof, the idea is that tumorigenesis is an instance when normal development goes awry, and cells keep on multiplying. That connection is difficult to test, however, because gene expression can’t be measured in the brains of living humans. Kho is working to show that the two processes are indeed parallel by comparing genes that appear to be related to human brain cancer to genes – in mice – that appear to be related to brain development.

Plugging the human and mouse data into his computer, Kho can tap a few keys and generate a three-dimensional cube that shows each point where a human and mouse gene turns on – each location corresponding to the signal the gene sends and when it is sent.

The analysis is complex, but the basic trend is easy to see: the genetic “beacons” seen at the earliest stage of brain development among mice cluster in roughly the same space as the “beacons” seen in tumorigenesis – suggesting that similar genetic instructions are involved in both processes. Just as important, the gene expression seen in late stages of brain development are largely missing in tumor development, giving more credibility to the idea that normal development includes an “off” switch that is missing in cancer. There is much more investigation to be done, but eventually Kho’s findings could bring other cancer researchers closer to treatment and prevention strategies.

Definitive diagnoses
Cancer is only one of many diseases with at least some genetic component; there are countless fields in which identifying genetic links could help clinicians diagnose disease sooner and more accurately, and even come up with new therapies targeting the specific genes responsible.

But it’s not as simple as it sounds. Single genetic anomalies rarely cause disease; instead, most diseases result from multiple genetic mutations and interactions. And before CHIP scientists can succeed, they need something they can’t derive from an equation: an enormous amount of patient data.

“You can’t study the genetics of a disease without patient data, and getting that data requires a lot of collaboration,” says Ingrid Holm, MD, an endocrinologist and geneticist at Children’s Genomics Center, which works closely with CHIP. “More clinicians are starting to get interested in the genetic side of disease now,” she says. “Children’s has done a lot of research into genetic diseases like muscular dystrophy, and is now starting to get into the study of the genetic factors responsible for diseases like congenital heart disease, asthma and allergy, autistic spectrum disorders, and diabetes.”

Holm helps those clinical researchers set up their studies and figure out what information they need to collect to facilitate useful genetic analysis. In addition, CHIP has developed numerous downloadable and Web-based tools to make it easier for researchers to integrate and interpret genetic information.

One of several clinicians currently working on a large-scale genomics study in collaboration with CHIP is Leonard Rappaport, MD, director of Children’s Developmental Medicine Center, and an expert in autistic spectrum disorders. Rappaport and Kohane hope to develop a genetic model for diagnosing autistic children, because, as Kohane puts it, “Even the best behavioral therapists are increasingly uncertain about their ability to characterize all the particular subclasses of the disorder.”

The notion of a single condition called autism has recently been overtaken by a more nuanced understanding of the problem. Today autism is viewed as an entire spectrum of disorders, ranging from severe to seemingly normal. By identifying patterns among the multiple genes associated with the disorder, researchers hope to define each disorder on the spectrum more clearly.

Autism is no longer regarded as a single developmental disorder, but a spectrum of disorders involving problems with social interaction and communication. Some autistic children function almost normally, while others lack a basic understanding of the world external to themselves.

“There’s strong empirical evidence that different types of autism may have specific prognoses and require specific interventions,” explains Rappaport. By associating specific genetic patterns (genotype) found in patients’ RNA with different autistic behavior and characteristics (phenotype), the collaborators aim to gain a new understanding of the disorder. “This study will help us identify the underlying biology of autism,” says Rappaport. “And will give us a more definitive way to diagnose individual children and tailor treatments specifically to their needs.”

Mapping public health patterns
It will take one to two years for Rappaport to collect enough patient information for a genomic analysis of autistic spectrum disorders. But another CHIP project is showing a real-world payoff right now. Kenneth Mandl, MD, MPH, an emergency medicine physician, CHIP investigator and research director of Children’s Biopreparedness Center, is using computer modeling to instantly detect unusual public health patterns.

In 2001 Mandl developed software that could detect unusual patterns in emergency room visits at Children’s. The system instantly compares the symptoms of new patients to a database of more than 500,000 emergency room visits over the past 11 years, and raises a red flag if it detects unusual activity. For example, if the system detects more respiratory symptoms than it predicts for a particular season and day of the week, physicians would be alerted to a possible virus outbreak.

Now, with funding from state and federal agencies, Mandl is developing a large-scale project called AEGIS (Automated Epidemiologic Geotemporal Integrated Surveillance) that will use data from multiple institutions across geographic regions. The system could allow public health officials to anticipate the trajectory of an infectious disease outbreak, identify environmental health problems such as contaminated groundwater, or catch the earliest signs of a biological weapons attack. The system can already predict an emergency room’s next-day volume within seven percent, but Mandl is continuing to expand the software’s capabilities by teaching it to interpret new types of information (such as lab results) to make even more sophisticated predictions.

“While the type of data this system sifts through is entirely different from the human genome, we’re using similar tools to what the geneticists are using,” says Mandl. “That’s because we’re both looking at how multiple systems interact and behave across time and space, and we’re both extracting patterns and clusters from this overwhelming amount of data,” he says.

At first glance, Mandl’s system may appear to have little in common with Rappaport’s autism study or Kho’s tumorigenesis investigations. But ultimately, they all wrestle with the same problem: modern medicine is generating new information at a pace far too fast for traditional analysis techniques to keep up with. Children’s Hospital’s Informatics Program is building the tools to help biologists and physicians clearly see the beacons that glow and recede in a sea of information, stretching to the horizons of their imagination. As the computing power of informatics grows stronger, deciphering the signals those beacons are sending out should become as simple as identifying blood cells through a microscope.

Software developed at Children’s can analyze patterns in emergency room admissions in real time, instantly detect unusual activity – such as an unforeseen spike in respiratory illness – and alert physicians and government officials about possible disease outbreaks and other public health threats.

To support research in the
Children’s Hospital Informatics Program,
contact Donna Richardson in the Children’s Hospital Trust
at (617) 355-2061 or donna.richardson@chtrust.org.