As high-throughput cellular analysis methods become more commonplace in research labs, biologists are struggling to sort through thousands of images at a time. The problem, according to Anne Carpenter, a postdoc at the Whitehead Institute, is that the software tools for analyzing these images have not kept pace with the imaging platforms that are generating them.
In an effort to break this bottleneck, Carpenter is developing a software package called CellProfiler that will be released under a GPL license some time this spring, when a paper on the project is slated for publication. Carpenter and her collaborators, David Sabatini, Polina Gollard, and Thouis Jones, presented a poster on the project at the CSBi (Computational and Systems Biology) Symposium at MIT Jan. 13-14.
Carpenter said that when she joined Sabatini’s lab just over a year ago, she conducted an “exhaustive survey” of commercial image analysis options to support the lab’s research on cell growth regulation using high-throughput microscopy and cell microarrays. This research relies on “subtle” phenotypes that standard image analysis packages tend to miss, she said.
“Commercial software is mostly written for pharmaceutical companies,” she said, “in which a defined visual output is measured. It was developed for phenotypes that are very straightforward and obvious to measure” — such as determining whether the cell had died or not, or identifying sharp changes in localization between the nucleus and cytoplasm, or even simply counting the number of cells.
But researchers at the Whitehead — and other academic labs — are increasingly trying to identify less obvious differences in cells, such as quantitative changes in particular proteins, or slight changes in localization, across thousands of images. This led Carpenter to develop CellProfiler, which can isolate individual objects, such as nuclei, from many images, while also providing several quantitative measurements for those objects.
One key to the software, according to a brochure about the project (available at http://web.wi.mit.edu/sabatini/pub/AnneWeb/CPbrochure.pdf), is an improved set of algorithms for segmenting objects — a stumbling block for many image analysis packages that have trouble separating clustered or overlapping images.
Once the objects — nuclei, cell boundaries, or components within the cytoplasm — are identified, CellProfiler offers several analysis modules for observing biological properties, including the number of objects, the size and shape of the objects, the intensity of the signal, and the texture of the objects. These properties, Carpenter said, allow researchers to measure many more types of phenotypes than currently available tools, “and not just answer yes-or-no questions.”
Carpenter said that the software is still undergoing beta testing among “a small group of academic and commercial collaborators.”
CellProfiler is envisioned as a complement to the software that is pre-packaged with imaging instruments, which is designed for “fast acquisition and fast image analysis, but the hangup is a lack of flexibility to analyze interesting phenotypes,” Carpenter said. The software’s open source license will enable labs to easily plug CellProfiler into their current image processing pipelines, she added.
Conversely, she said, the Whitehead team is offering commercial software providers the option to contribute modules for their packages as “plug-ins” to CellProfiler. These firms would still maintain the IP for their tools, she said, but the arrangement would give users access to a broader range of software options.
Once CellProfiler is released (at http://www.cellprofiler.org), Carpenter said she envisions the website serving as a repository for image analysis modules submitted by the community. “There are great algorithms and bits of software in the literature — in the computer vision research field, especially — that never make it to being used by real biologists,” she said. The CellProfiler infrastructure could ultimately serve as a resource for biologists to find the best analysis modules for different image analysis tasks.
Carpenter and her colleagues are also collaborating with members of the Open Microscopy Environment (http://www.openmicroscopy.org/), a collaborative project to develop a system for storing image data, metadata, and analytical results. The OME developers are now looking into integrating analytical tools with the data management system, and Carpenter said that CellProfiler would be among the first of these tools included in the platform.
The two open source image informatics projects “are really complementary,” she said, “and come at the perfect time for each other.”
Further information about CellProfiler is available at http://people.csail.mit.edu/u/a/anniebio/public_html/CellProfiler/.