Entry 2 – The CyTOF users group meeting in Cambridge
I spent a day in November at the CyTOF users group meeting in Cambridge, Massachusetts, and if this blog will touch on technology, what a great place to begin! The CyTOF is a mass cytometer produced by DVS Sciences. The name is an abbreviation for ‘cytometry Time Of Flight’, and the machine is used to conduct cytometry experiments on a much larger scale than can be achieved with a traditional fluorescence based cytometer.
In a mass cytometry experiment instead of tagging antibodies with fluorescent molecules, the tags are isotopes of any element that is not found naturally in the biological system under study. Cells labeled with these antibody-isotope conjugates are then vaporized, ionizing the elements. This is done within a magnetic field so that the particles move across the magnetic field to a detector at a rate proportional to the size of the particle; the lighter the particle, the faster it moves. The TOF part of CyTOF refers to the time it takes for these particles to fly to the detector. What makes this process so useful is that there is very little noise associated with the ion tags. Each element can be resolved so no compensation is necessary and any element that won’t occur in the sample is theoretically in play to be used as a reagent. DVS advertises that there are over 100 detector channels. Think about what that could do for your ability to characterize the immune system, or create a disease profile!
There are some folks who are thinking about just that. The users’ group meeting featured two really nice application talks. Dr. Bill O’Gorman from Stanford used the CyTOF to profile the immune response to influenza vaccination, while Dr. Jonathan Irish of Vanderbilt focused on cell signaling. And while I am just a simple unfrozen-caveman biomedical engineer, unable to understand the complexities of these systems, there is one thing I do know: I ran a lot of 4-color experiments that failed to explain satisfactorily what was happening in the biological system. The body has so much redundancy that focusing on any one signaling pathway isn’t enough. This tool looks like the next step in the evolution of the cytometer.
I’m in on the technology, and have written some documentation on how to use CyTOF data in FlowJo. But my take-home message from the meeting, as someone who deals with data analysis, was that the analysis tools left something lacking. Many of the presenters showed some variant of a graphic that was essentially a multi-graph overlay from FlowJo for a 30-parameter experiment. The screen filled with a host of 2D plots, which presented a daunting challenge to understand holistically. While preprocessing and organization can help make this more manageable, considering that the experiment could grow to 100-parameters in the not-to-distant future makes me believe that some algorithmic analysis will be needed for this kind of work.
The algorithm du jour is SPADE – spanning tree density estimation. But SPADE is not the be all and end all. It replaced Principle Component Analysis (PCA), and already there are people who believe a tool called VISNE will take SPADEs place. In the end most people who work in the field of automated analysis conclude that there isn’t a single best algorithm for every job; depending on the project a tool might fare better or worse. In fact, my friend Dr. Ryan Brinkman of the BCCRC, perhaps the leading authority on automated analysis in flow cytometry, is of the opinion that using multiple algorithms in an ensemble approach is best.
The puzzle I’m working on now is ‘how should Treestar make analysis of this cool new technology easier’? I think the answer is to give the people what they want, attempt to keep adding tools and plug-in ability, and try to stay ahead of the curve by keeping an eye on the leading edge. What people want right now is SPADE, so we’ve put it in FlowJo, and I’ve written documentation on how to use SPADE in FlowJo. We’re adding some general R-based clustering tools (R is an open source programming language). I’d be real happy to hear from any users regarding what algorithmic data analysis tools are most valuable to them. I’ve put up a page of documentation on how we are incorporating R-based tools into FlowJo, if you would like more information on that.
While my take home message from the user group meeting was that there was work to do, the feeling I walked out with was excitement. I can imagine that this new tool is going to open a lot of possibilities for your research, my bounds on playing with algorithms, and humankind’s knowledge of how we work.