User:Steven J. Koch/Notebook/Kochlab/2010/04/15/Gliding assay analysis with Larry and Andy

From OpenWetWare
< User:Steven J. Koch‎ | Notebook‎ | Kochlab‎ | 2010‎ | 04‎ | 15
Jump to: navigation, search

Steve Koch 01:11, 16 April 2010 (EDT): Over the past couple weeks, Larry, Andy and I have been looking at Andy's gliding assay data using Larry's tracking software. The two of them have perfected their aspects of the project to amazingly high-quality. So, we're now at the point where we have TONS of data and are figuring out how best to analyze it. We've been writing some quick code here and there, and undoubtedly will add some features to the tracking software. It's become obvious that we do not have a good way of capturing what we're doing during this data analysis trials that we're doing. It will be easy to capture the final analysis once we know what we want to do. But at this point, we're learning all kinds of things about the data but we don't have time to summarize it in our notebooks. Larry and Andy will be working on ways to make capturing this easier. For example, automatic screen shots saved at appropriate times that we can easily link to.

Larry -- We currently save the data for all tracking attempts, which was a slick feature we added and helped us today. I think all those should be public. We currently can't make all the raw data public because we don't have enough space on the server and I can't figure out how to make Drobo public. Maybe you can modify the tracking software to save additional copies (along with the added screenshot) to the webpub directory? As long as they have dates and times, they don't need to be organized. Thanks!
  • I have manually added all of our tracking results to the following location. Anyone can access any of it, but it's probably not going to be meaningful. But hopefully below I will link to specific files.

For now, I'll summarize where we're at now:

  • We think the best way to report the data now, as far as speed changes go is to:
    • Estimate the probability density function for instantaneous speeds (using a kernel density estimator--similar to our current "sliding window histogram"). Take the most likely value in this pdf as the "gliding speed." Any lower peaks we will ignore for now.
      • If the most likely peak is for a lower speed peak and there is clearly a bimodal (or more) behavior, then we will either use the higher peak as the "gliding speed" or we will throw out the data set.
  • For a while, we thought that even good MTs that seem to glide well were switching between two close speeds. However, Larry modified his image simulation program to move the MTs according to poisson statistics. When tracking these data sets, we also saw bimodal behavior. The image simulation software is brilliant and has become essential for understanding the real data. We learned that two closely-spaced peaks can arise commonly, even when there is only one underlying average rate. These peaks are an artifact and the speeds corresponding to the peaks tend to depend heavily on the smoothing window size. In contrast "real" bimodal behavior is less sensitive to window smoothing size, it seems. Although we haven't proven this yet.
  • we found one good MT that seemed by eye to definitely stick momentarily.

Stalled MT April 15.PNG

  • By eye it looks like clearly the MT stalls. However, the analysis of smoothed data only shows dipping down to 300 or so. This must be an artifact of smoothing. I still think we're smoothing correctly, but it's surprising that it doesn't get closer to zero. We'll need Larry's image simulation program to settle the issue for sure.