This week, I stomped through the swamp, rolled the boulder, weeded the strawberries. In science, sometimes we get an epiphany but other times just pith. I don’t mean as in a ‘pithy remark’ but as in the featureless, spongy tissue in the middle of a stem.
I spent another two hours imaging cell walls on these most challenging samples. Many more hours spent putting the cell wall images through analysis. This sounds like they are on a couch and I am asking them about their mother. But no. Analysis as in quantification.
That I have to quantify should be no surprise. As I have been writing about here, Joseph Hill has been sending me sections of plant stems and I have been kvetching about my struggle to get consistently good images. But I have also said how the sections are from five different kinds of plants: the wild type and four different mutants. Strictly speaking, three of the plants are single mutants and the fourth is a triple mutant combining all three. My job is to determine how, or if at all, these plants differ in the structure of their cell walls.
All along, I have been looking at these images. Back in the office after a two-hour session on the microscope, it is a pleasure, usually, to take ten minutes to scroll through the loot. Like getting back to the hotel after a day spent exploring and looking at all the photos you took. My eye and brain have their impressions. One of the plants, in fact the triple mutant, has cell walls that are so strikingly weird that it took me several tries to convince myself that what I was imaging was actually cell wall (see here and here). With images like that, personal impressions go fairly far. But even here there is a problem. The journal will let me represent the cell walls of these plants with one or at best two images each. Which one(s) should I pick? As I have mentioned before, various images of the same kind of cell wall (i.e., from the individual cells of a given kind of plant) vary quite a lot. How do I chose a representative image? What does that even mean? With all that variation, in what way can a single image be properly representative?
Even so, with a cell wall structure as weird as I see for the triple mutant, I could probably get away with showing a picture and stating the scientific equivalent of “Holy crap!”. It is that unusual. But what of the other three mutants? These look certainly less weird than the triple and sort of different than the wild type. But “sort of” is precisely what won’t fly in a publication. Obviously – it is frustratingly vague. The point of analysis is to replace “sort of” with something measured: The cell wall of plant A differs from that of plant B by x units of parameter z. Bang! That’s how they differ.
The above is probably freakishly obvious – scientists measure stuff, its what we do. If you have been anywhere near this kind of thing before you are probably already thinking about reproducibility and statistical tests, way ahead of me here. That’s fine. But for a microscopist, it is not always easy to figure how to measure something. Imagine you have pictures of granite blocks from two different quarries and you want to know how (or if) the minerals differ (Fig. 1). Well, why not measure the size of the flecks? Sounds good until you try it. Flecks are irregular and finding their length or width seems highly arbitrary. OK – how about the fleck area? Again sounds great until you try it and see that one fleck grades into another and the boundaries are evasive and difficult to pin down well enough to draw a ruler around. And I haven’t even mentioned color. Non trivial problem.
![Figure 1. Fleck-tones. Examples of granite. They look different by eye but how to measure? Image on the left is from http://geology.com/rocks/granite.shtml. On the right from http://www.armoireconversion.com/granite.html](https://websites.umass.edu/baskin/files/2016/10/granite-granite-600-1024x341.jpg)
Some years ago, I invented an analytical process for measuring things in images of cell walls. It is not as long and drawn out as psychoanalysis but it is drawn out all the same. Although the heart of thing is a piece of software that takes about 10 sec to run, everything else is manual. If I were a coder I could probably automate the whole schmear but for now I have to do it by hand. The most time consuming part is taking the numbers that come from each image (the output of the software routine) and melding them into something useful. In a future post, I’ll lay it all out because it will be more fun when there are conclusions, however tenuous. My point here is to recount that I spent some hours pushing images through this analysis. It was the science I did this week. But as I titled the post, it is in the middle of it all so I was working mostly like an automaton: cutting and pasting, making plots, averaging numbers. In due time, I hope to pull a rabbit out of this hat. But right now it feel more like hocus pocus.