A disclaimer to start with: This comment does not claim to be correct in any way. Please feel free to enjoy all typo's, spelling errors, omissions and inconsistencies with previous comments. If I made them before my last coffee break I would have forgotten anyhow. I liked Ray's and Mario's approach to show a few examples. It shows the artistic freedom one can get on how to display the data. And they only used one data set and two software packages and far from every possible iteration of scaling smoothing and colour. And they happily accepted Fluor:CD8 as an axis label. I would have thought it to be fluorescent if that is what Fluor means. I tend to write for example CD8*FITC 525nm log which still doesn't give you bandwidth etc but tells you at least it is green fluorescence. I actually love density plots (please don't call them colour dot plots as this tends to be used for colour gated dot plots) over contours and actually most of all a 3-D display where I can see mountains or clouds. Unfortunately we can not yet print the moving pictures described in Harry Potter but that is when the fun really starts. However, it can be done in on-line publications. The confusion created by our beloved contour or dot plots can be enormous. Not for no reason one of Marc Abraham's AIR-logo's shows a dot plot under the stinker (http://www.improbable.com/bookstore/bookstore-top.html). Actually nicely shown in Mario's example of smoothed contour plots is how you can make cells vanish. Are these small populations between the single and double positive cells important? or perhaps a compensation error?... If one wants to show those cells to make a point, one would not use the smoothed contour plot. If one was to sort them, the contours would lead nowhere. From that point the old Ortho 50H software with it's grey level dot plots was far advanced at it's time. With 16 grey levels it allowed a huge dynamic range for density plots (e.g. 1=2^0 to 32768=2^15 per channel) w/o compromising the resolution at the bottom - unless you do not like the log-2 z-scale. Let's step back a bit Why do we publish? Normally to convince someone else about the importance and values of our work. Why would we add those funny little pictures? To convince the reader with a visual aid about our data, the population we have seen doing something, e.g. once you seen a cluster - you shall believe. They are descriptive sketches and nothing else as they don't give is any statistics themselves. In particular with low event numbers it becomes very important to reflect on confidence intervals of your data as nicely presented by Terry Hoy on the Royal Microscopical Society Flow Course is Sheffield a couple of weeks ago. However, the plots have something in common with statistic: Statistics are like a bikini - what they reveal is suggestive, but what they conceal is vital. (Aaron Levenstein) Now if we argue the use of plots for illustration only but the use of "solid numbers" to substantiate our work with "sound statistics" we have to keep in our mind that all this number crunching only characterises / validate the position of the dots on the screen, nothing else. We can only test the numerics of our measurements and not the measurement itself. Apart from the requirement for correct instrument set-up this is subject to scientific reasoning; about how the measurement was performed and what controls were done, if saturation was reached etc. in order to interpret the indirect measures we have undertaken. This includes detailed description of clones and concentrations and control data and "unmassaged" pictures of raw data and gated data. The question remaining is how to validate the quality of the data and who is to do it. For validation in house for example we display / print all fluorescence channels versus log side scatter ungated. For whole blood this helps as it gives you already a differentiation of lymph's, mono's and neutrophils (and eosinophils if for example using FACSlyse). Thus if you suddenly spot CD8 positive neutrophils you know that there is trouble. For cultured cell lines log forward scatter vs. fluorescence seems to be better as in a lot of cases the cells with lower forward scatter are the ones on their way out. This type of display also indicates the presence of fluorescent aggregates that can lead to problems by sticking to or coinciding with cells (how do we want to report coincidence factors?) and shows if the antigen expression is scatter / volume related or not. But this is all a matter of personal taste. I do not believe we can eradicate poor publications. On one side there is always human error. On the other side there is an incredible inflation of science and in particular the numbers of publications and journals, inevitably leading to a decay in quality. We can only try to educate people in their ability to set up experiments and analyse data. Attempts like Mike's CD-ROM approach or Paul's virtual laboratory where you can even do the virtual pipetting of antibody are leading in the right direction. The latter could even be spiked with pitfalls like changing the buffer batch in an experiment, picking one made up with water contaminated with red fluorescent algae. Now in a proper clinical setting, as I know from my brother, all the results have to be signed off and interpreted by him or a colleague as a specialist before they leave the lab for the physicians out in the University. Unless you run your samples in a core facility with an experienced cytometrist, such gate keepers may not exist. For political financial and logistic reasons a lot of people have bought their own instruments independent from an already existing expertise. The time and effort put into the proper planning and analysis of experiments by someone who knows his cytometry would cost (not money necessarily). If the gate keeper is to be the reviewer of an article then he should not only be skilled enough to understand the flow data but he also would require some validation information even if not published as it is not possible to judge complex work from one or two pictures. In that context I think that the suggested discussion board for post publishing review offered via the publisher to their subscribers speaks in favour of the publisher, as this type of feedback does allow him to improve the journal quality. I am sure we can come up with a suggested list of information that should be included on cytometry experiments if applicable and make it easier not to forget the necessary controls and I would ask James Watson (http://www.cyto.purdue.edu/flowcyt/books/bookl.htm) to join the advisory party that might be set up for that. But there will always be a lot of poor data published, for numerous reasons, and the only one to blame is the reader that believes information to be true because it was printed in a newspaper or a publication to be correct just because it was published in a scientific journal (see www.improbable.com for further enjoyment on that). Gerhard
This archive was generated by hypermail 2b29 : Sun Jan 05 2003 - 19:01:36 EST