My actual research has nothing to do with behavior, which could be a bit of a surprise given that all I've written about on here (so far) is my reaction to a popular science book about human behavior. In reality, I study the process of evolution itself, and I do so using bacteria. Bacteria offer a number of useful advantages in studying evolution: rapid generations, easy to manipulate environmental conditions, large population sizes, few ethical problems (things without neurons can't really suffer the same way tings with brains can), and the ability to freeze the population and revive it later.
With the sort of work I do, one thing I frequently need to measure is how many bacteria of a given type I have in a particular flask at a particular time. Population sizes, as I've said, tend to be very large in bacteria -- the populations I work with are typically around 3 * 10^8 cells -- and it's not really reasonable to count them all. It would also sometimes make the experiment useless, because the cells I take out to count aren't then still in the flask doing whatever it is I want to measure them doing. Instead, what I want is to use a small fraction of the population to estimate what the full population is like. I want a random sample from that population, which is pretty easy because I grow things in liquid culture and my cells aren't capable of making biofilms, so I just need to use a vortexer and make sure the liquid's mixed. Then I need to work with just a small amount of this liquid, and somehow get a reliable count of the number of cells in it.
So far, I've used the word reliable. This is because I'm a bit careful about terminology, and in science, accurate and precise have defined meanings. Ideally, I want my count to be both accurate and precise. Accurate refers to the sample number being close to the real number. Precise refers to multiple different measurements being close to each other. I made my own version of this common figure to demonstrate that, since I couldn't find one that I was sure was free to use (also, so I could ensure it was readable under different color vision impairments):
(Feel free to use this image if it's not for commercial purposes, but if you're going to just copy it I'd appreciate you attaching my name: Mike Wiser)
If the goal is to hit the center of the target, the two images on the top are accurate: if you take the average of all the shots, it will be close to the center. The two images on the left are precise: there is little scatter from one shot to the next. The lower right is neither accurate nor precise; there is a lot of scatter between the shots, and the average of all of them isn't very close to the center of the target. Scientifically, it's fairly easy to test for precision, but it's harder to know much about accuracy in a measurement.
This brings me to the paper I wanted to bring up. When I take my sample of liquid, I then dilute it and spread some of the dilution on an agar plate. Agar is basically science gelatin -- it makes the growth medium thicken and become a semisolid, rather than sloshing around as a liquid. I aim to get a dilution that has a high enough population size that I don't have so few colonies that the variation from one plate to another is larger. But I also don't want to have too many colonies per plate, since colonies can grow into each other (reducing accuracy) and because really dense plates take longer to count (and my own reliability may go down as a function of fatigue). It's long been the lore in the labs I've worked in that you should aim for between 30 and 300 colonies on a plate for reliability; whether that is accurate or precise or both is another matter. That range gives plates such as:
(Apologies to those with color vision limitations -- my work involves me counting the red colonies, and counting the pink colonies, and I realize the distinctions are not terribly visible in certain forms of color vision limitation. As above, if you'd like to use this image and it's not for commercial purposes, feel free so long as you credit my name to it: Mike Wiser)
That seems like a reasonable number of colonies on the plate. But where did this lore come from? A recent conversation I had with a labmate (Alita Burmeister) brought up that she had heard from one of the professors (formerly) down the hall from us that this came from an old study from the dairy industry. With the professor's (Tom Schmidt, now at the University of Michigan) help -- both his, and one of his collaborators (Clive Waldron), I found the paper I was looking for: Breed, Robert S and W. D. Dotterrer. The Number of Colonies Allowable on Satisfactory Agar Plates. J Bacteriol. 1916, 1(3):321-331. (full paper here) In it, the authors spread various dilutions of milk on agar plates, and counted three plates for each dilution to look for ranges in which each of the three counted plates was no more than 20% different from the average. Plates were counted after both 5 and 7 days. As expected, there is an intermediate range of colonies per plate that results in fewer discrepancies between plates than are found in plates with either too few or too many colonies. What exactly this rate of discrepancy is changes from 5 to 7 days, but the general finding is robust. To quote the authors, "Plates having less than 30 colonies or more than 400 colonies show very large percentages of discrepancies." Further, the type of discrepancy changes; plates with an average of 50 or fewer colonies tend to have discrepancies from one or more plates having more than 20% more than the average, while plates with an average of 200 or more colonies have discrepancies from one or more plates have more than 20% fewer colonies than the average.
I think it can be interesting at times to track down these old bits of lore to see who actually did an experiment or made an observation that became a standard part of normal practice within a lab.
No comments:
Post a Comment