BNMC Imaging Meeting

participants

Meeting Goals

On February 2, the BNMC organized a small meeting of Caltech people involved in image processing in biology. The goal was to let participants meet and learn from other Caltech researchers who may be working on similar problems, and to help the BNMC shape its goals with respect to image processing in biology.

Summary of Meeting Presentations

Mike Hucka opened the meeting with a brief introduction focused on the Beckman Institute's BNMC itself. [Slides available in PDF format.] The BNMC is a collaboration between multiple Caltech groups and individuals, and its aim is to foster the sharing of experience and infrastructure, with the ultimate goal of bringing together biology, theory and computation in the service of research in biology. The BNMC is jump-starting its research efforts by using existing engagements in the areas of developmental modeling in plants (especially the Computable Plant project), software infrastructure for systems biology (especially SBML, BioModels.net, SBGN), and multiscale stochastic modeling (with Dan Gillespie, Linda Petzold, John Doyle, CACR, and others).

meyerowitz

Elliot Meyerowitz. In their imaging work, they had started out using the imaging setup from Scott Fraser's BIC originally, but later developed their own software and facilities. They label shoot cells in the growing tip of _Arabidopsis thaliana using a green fluorescent cell membrane protein, then perform time-lapse imaging to extract relative gene expressions from 2-D images. Their confocal imaging setup places a high-end microscope right up against the plant tip, submerged in water. For alignment of the confocal images, they use a combination of commercial packages and custom software. They have developed tools for tracking nuclei positions, by calculating a vector velocity field and analyzing this. The nuclei tracking work was done as a collaboration with Eric Mjolsness, Tigran Bacarian, and Victoria Gor. An Amira employee has also developed an Amira module for them for calculating and displaying the tensor field data for growth analysis. Marcus mentioned some items on his wish list:

Al Barr talked about the Beckman Institute pilot project, Automated High Throughput Behavior Capture and Screening. The project's purpose is to develop new technology and a prototype system for screening complex organismal phenotypes related to animal motion and behavior modes that are dominated by motion. They are approaching this by developing new hardware and software modules whose combination autonomously tracks and processes the motion and behavior of individuals within controlled environments. The ultimate system will consist of a set of image-capture and automated model-making tools based on "generative" modeling and feature extraction that will permit the high-throughput quantification of traits that manifest as a change in body motion, orientation, or shape. The current pilot project is a minimalist proof-of-concept prototype. The ultimate goal is motion capture including markerless tracking. Al Barr also presented some visualization results based on Kurt Fleisher's Cellular Texture Generation (ACM Siggraph,1995). Al listed some of the items on his biological software wishlist:

A problem solving environment for behavioral analysis. Suggestions for developers:

  1. Use a bottom-up approach

  2. Make it compatible with existing environments such as Mathematica

  3. Specialized hardware is good, but needs to be adapted so that it is completely transparent to biologists users

  4. Cross-talk and coupling between different chemical pathways and systems, such as EGF and Notch, with Paul Sternberg.

Grant Jensen described imaging with electron tomography utilizing a 300 kV cryo transmission electron microscope (TEM). Image processing in their approach involves three dimensional reconstruction of cells by panning through a sequence of 2D images. They have developed a large collection of Amira modules developed in-house, as well as additional in-house software. They also have experience with Bsoft, Peach, IMOD, Priism, PFT3DR and Dino. His wish list includes systems biology software that can incorporate molecular structure, and he would like to collaborate with developers by providing data for systems such as HIV virus self-assembly, chromosomal rearrangement, bacterial division, and time-courses of every molecular is particular systems.

Edoardo Marcora of the Kennedy lab showed how they reconstruct 3-D images of dendritic spines to determine morphological properties such as shape, volume, density. The (public and free) software that he uses is cumbersome and he would like something better. In the future, he would like to fully image signalling pathways from the synapse to the nucleus, by photolabeling synaptic proteins and performing synaptic photostimulation. His wish list includes:

Sean Megason and Michael Liebling talked about projects in the Caltech Biological Imaging Center (BIC), managed by Scott Fraser. Sean Megason discussed the digital fish where they are performing in-toto imaging of complete lineages of the entire organism. Every cell in an embryonic zebra fish is labeled using two-color GFP's, and every cell is imaged through space and time (4D data). In particular, he showed z-stack time histories of inner ear formation. They then use software to extract, analyze, and integrate data. They have developed the imaging analysis software package GoFigure (MS Windows only) which is specifically designed for visualizing, tracking, and analyzing cells in multidimensional confocal/two-photon image sets, including time-lapse series of confocal image stacks (xyzt), with automatic segmentation and tracking cells in 2D, 2D+time, 3D, 3D+time, and 3D+time+cell division. The system uses a MySql database back-end so it can handle extremely large image sets. They also use the Visualization Toolkit (VTK), an open-source C++ class library for computer graphics, and the NLM Insight Segmentation and Registration Toolkit (ITK). Both packages are TCL and Java scriptable. Michael Liebling discussed image reconstruction of the embryonic zebrafish heart, integrating existing tools into commercial software, primarily ImarisXT-MATLAB. Michael also referred us to the OME XML Schemas, a developing standard for microscopy data. Their wish list includes:

Changhuei Yang described research in the Caltech Biophotonics Laboratory. In this lab, they are using optical coherence tomagraphy (OCT), which has similarities to ultrasound but works in the light spectrum. They have developed a forward imaging probe that is small enough to be usable for clinical biopsies. They also use optofluidic microscopy, for example to image organisms like C. elegans at 600 nM resolution. Current challenges faced by the lab include developing volumetiric image representations that are useful for surgeons in real time, sequential collation of images, and rescaling and Fourier Transforms in real time. Future challenges include developing a multi-spectral imaging system as well as ultra-high-resolution imaging systems at 30nM resolution.

Mory Gharib described the Center for Quantitative Visualization. They use Digital Particle Image Velocimetry to extract cell movement from biological images such as those obtained from the BIC. In particle image velocimetry, they obtain instantaneous global measurements of the entire fluid and use cross-correlation between sequential images to determine the velocity field (DPIV). In DPITV, they also measure temperature simultaneously; in DUSIV, they perform digial ultrasound speckle image velocimetry; and in DDPIV, they can extract forces from the curl of the velocity field.

Michael Elowitz described research into how genetic circuits, composed of interacting genes and proteins, enable individual cells to make decisions, oscillate, and communicate with one another. Systems include both synthetic (e.g., the repressilator) and natural circuits (e.g., in bacteria, yeast, and mammalian cell cultures). He utilizes multi-color time lapse imaging of 3 or more simultaneous GFP probes. This work relies heavily on computational analysis, particularly in MATLAB. His wish list includes:

Eric Mjolsness and Tigran Bacarian talked about image processing and classification problems and software that they have developed and applied to problems in the Computable Plant with Elliot Meyerowitz and the Elowitz Lab. The correspondance problem is the problem of determining which elements of an image at two different times correspond to one another, and they have developed software to address this problem using their softassign algorithm which they utilize for tracking moving cells labeled with GFP.

Discussion

During the break and at lunch, the group continued with a free discussion about the current state of image processing at Caltech and in their research applications. A number of needs emerged:

doyle

Regarding multiscale issues:

One of the topics of discussion was people's desire for a general purpose problem solving environment where they could perform image analysis tasks. But there are many challenges in addressing this need. Some of the issues brought up and conclusions reached at the meeting:

bnmc banner