November 2007 - August 2012
Hanchuan Peng develops bioimage analysis and informatics techniques. He uses these techniques to mine and fuse knowledge from three-dimensional animal brain images, at both micrometer and nanometer scales. His group is building 3D neuronal atlases of brains – incorporating neuron distribution, projection, and connection statistics and mapping functional data of neurons.
He uses these techniques to mine and fuse knowledge from three-dimensional animal brain images, at both micrometer and nanometer scales. His group is building 3D neuronal atlases of brains – incorporating neuron distribution, projection, and connection statistics and mapping functional data of neurons.
How does a brain work? To answer this question we will need a map, or "atlas," of a brain to help us explore its anatomy and neuronal wiring and to associate neuronal events, such as the firing patterns of neurons, with animal behaviors.
Such an atlas, in its digital form, can dramatically improve our ability to tackle the secrets of a brain, from the level of a single neuron or synapse to that of the entire brain. Recent advances in multicolor labeling and high-resolution imaging have made it possible to reconstruct computationally the 3D digital neuronal atlas of an animal at the single-neuron, whole-brain level.
In collaboration with several laboratories at Janelia, we are developing novel image informatics tools to build a multiscale 3D digital atlas for the brain of a fruit fly (Drosophila melanogaster), a widely used model system in biology. A fly brain is estimated to have 100,000-150,000 neurons. In a simplified view, the neuronal network in a brain can be described as a forest, with each tree representing a neuron. It is crucial to develop data-mining techniques to identify, describe, compare, categorize, and search these trees and branches, and to model computationally the distribution and interaction of objects in the context of the entire forest. By targeting the fly brain, our goal, which is shared with other Janelia labs, is to build a high-resolution 3D digital atlas of a fly brain that includes the statistics of neuronal distributions, projections, and connections. To do this, we will use bioimage informatics and mining tools developed through collaboration.
One aspect of our research aims at bridging the gap between the micrometer- and nanometer-resolution image information collected through both light and electron microscopy. In previous studies, nanometer-scale images were often used to study connections of neurons. Micrometer-scale images were usually considered when comparing gene expression patterns and investigating neuronal functions. Appropriate integration of these two sources of data can significantly deepen our insights into how neurons distribute, connect to each other, and function in a brain.
Despite many accessible techniques for images acquired at a single-resolution scale, we still lack a set of high-performance techniques and software tools for multiscale bioimage mining and informatics. Our goal is to fuse and transform the enormous volume of heterogeneous information acquired at different imaging scales into meaningful knowledge. To do this, we are building a suite of tools of image analysis and mining, as well as visualization.
As a prerequisite to integrate information of different bioimaging scales and modalities, we are working on several techniques for analyzing brain images and neuronal patterns.
- We have developed and continue to improve 3D image registration software to align fly brain and thorax images, which could differ significantly in their morphology, intensity, orientation, and resolution (scale), and often correspond to different brain regions or have various artifacts.
- We are studying how to automate the extraction, or tracing, of neurons from 3D microscopic images and how to characterize their morphological and topological features.
- For registered brain images and reconstructed neurons, we will develop large-scale, high-throughput data-mining techniques to search, cluster, and classify patterns and detect the associations between these patterns and functional variables, such as animal behaviors. We are also developing computational tools to help biologists compare, annotate, and measure brain images.
3D Digital Atlases of Animals at Single-Cell (Neuron) Resolution
In a recent collaboration with Eugene Myers and Stuart Kim (Stanford University), we have developed a series of computational approaches to produce a 3D digital cell atlas for the animal Caenorhabditis elegans, and thus have collected the statistics of gene expression patterns at the single-cell resolution.
From an image-informatics perspective, building the 3D statistical neuronal atlas in a fly brain is conceptually similar to, although technically more challenging than, building the digital cell atlas of C. elegans. It is generally believed that many neurons in a fly brain are stereotyped. Collaborating with other Janelia labs, we are developing algorithms and computer programs to collect the statistics from many fly brains at different resolutions. This will result in a 3D statistical atlas, containing the statistics of neuron spatial distributions, projections, and connections.
We are also interested in using these statistics to (1) model the wiring of neurons in a fly brain, (2) detect the associations between neuronal distribution/connections and the animal behaviors, and (3) map other neuronal activity data (e.g., firing) to this atlas and analyze the relationships among these physiological neuronal activities and animal behaviors.
Machine Learning, Signal/Image Processing for Bioimaging
My lab is also developing new machine-learning and signal/image-processing methods for bioimaging data. We are interested in designing and implementing computational methods to promote various aspects of high-resolution imaging and improve our ability to understand microscopy images, including those related to super- spatial and temporal resolution and deep-tissue imaging. These research projects are natural extensions of my previous studies in areas such as feature/model learning for biomedical and multimedia signal/images. We are also interested in using these new developments to tackle problems in related domains of computational biology, e.g., to understand gene expression patterns and underlying genetic regulatory networks.