For centuries, anatomists have drawn illustrations of the nerve pathways, skeletons, circulatory systems, and internal organs of the body to help them figure out how the different parts work together.
Today, computers, microscopy, and molecular genetics allow scientists to build more sophisticated, three-dimensional representations of life and its biochemical processes. But to "sculpt" such models, researchers must synthesize information from thousands of detailed photographic images that can show different magnifications of a structure or different biological characteristics, such as varying gene expression across cell types. Each cell in the body has the same genes, but a cell's identity depends on which genes it expresses.
Developing computational tools to facilitate analysis of massive amounts of biomedical images and their associated data is the scientific mission of biomedical engineer Hanchuan Peng. "Biologists want to be able to analyze meaningful patterns in the many images they generate from varying experimental conditions, but are unable to do so manually," Peng says. "We take advantage of advanced computing technologies to help them," he said.
Peng is part of a scientific team at HHMI's Janelia Farm Research Campus that is developing a searchable, high-resolution, 3-D atlas of the brain of the model organism Drosophila melanogaster, also known as the fruit fly. Much is known about the genes implicated in fly behaviors and the brain locations where the behaviors originate. But the scientists at Janelia hope a digital brain that integrates large regions, such as the optic and antennal lobes; smaller structures, such as the connections between its 100,000 or so neurons; and experimental data, such as patterns of neuronal firing and gene expression, will allow them to discover even more.
Ever since Peng was a graduate student in biomedical engineering, he has been designing computational tools that address, in some way, both imagery and neurobiology. For his Ph.D. thesis, he studied artificial neural networks, so named because they simulate biological neuronal connections. With his models, Peng wrote a computer program that recognized handwritten digits and won him a first-place award in a national competition of software development in China.
But Peng wanted to learn more about the real brain. So in 2000, he went to Johns Hopkins University for a postdoctoral fellowship to work on a project using MRIs (magnetic resonance images) to assess mild cognitive impairment, a condition believed to be a precursor of Alzheimer's disease.
The researchers at Johns Hopkins had followed a group of elderly individuals over several years and had monitored their learning or memory capabilities. They also had taken a series of MRIs of the seniors' brains. Peng designed a computer program that analyzed the MRIs for structural changes of the brain related to symptoms of decreasing cognitive capability. Ultimately, the team identified new brain regions associated with mild cognitive impairment, a finding that could lead to earlier diagnosis of the condition.
Peng's subsequent postdoctoral fellowship at Lawrence Berkeley National Laboratory led him ultimately to Janelia. While at the Berkeley Lab, he once worked with Gene Myers, who would later go to Janelia. At Berkeley, Peng and Myers developed a computational method to analyze a database of thousands of images of the developing fly embryos. Their tool allowed them to search the images and find genes expressed at the same time and in the same region of the embryo. They also worked with Michael Eisen, a Berkeley geneticist, developing a new approach to computationally reconstruct developing gene expression patterns, without live animal imaging.
"At the time [2002–2005], computational biology was becoming very hot and everybody was trying to find future directions in the field," Peng says. "I was lucky to find a good direction, which was to deal with large-scale bioimaging analysis." Image analysis has become very important for modern biology, which generates many data-laden pictures. Indeed, the annual bioimage informatics workshop, which Peng started in 2005 at Stanford University, is now highly attended.
At Janelia, Peng, who came in 2006, and Myers work together on the brain atlas with Gerald Rubin (Janelia Farm director), Julie Simpson, and others. Peng and Myers focus on computational issues, while other biologists and physicists perform experiments with flies and generate images related to brain gene expression.
The challenges of developing the atlas are numerous, Peng says. The brain, for example, has many types of neurons, each of which has to be "idealized" for the model. The connections between the neurons also vary throughout the brain. "A single computer algorithm might not be able to deal with a different neuronal structure in a different location in the brain," Peng says. "We are struggling to develop a spectrum of computational methods to handle these problems."
His colleagues also take photomicrographs at different magnifications. Electron microscopy can show synaptic connections of neurons at the nanometer level, while light microscopy reveals the more global picture of neuronal structures at the micrometer level. Peng is developing computational methods to integrate the data from the different magnifications into the 3-D brain atlas.
Peng is confident the collaborative nature of the fly brain project will lead to success. He also expects that the platforms he develops will be applicable to other organs in the body and for other types of imaging analysis. "It is exciting to have this opportunity to work with all these people," Peng says. "Although there are very smart people at Janelia, no one can do everything, and we depend on each other. Together we can do fantastic things."