The experimental plan included 2 groups with 8 cell lines in each group. For each cell line 10 cells were chosen randomly and at least 300 vesicles were tracked in each cell.
I now want to compare the 2 groups to evaluate the vesicle speed and distance travelled.
I am unsure which statistical test needs to be applied to compare the 2 groups considering differences in age, sex and related mutation of the cell lines.
Group 1 (Control):
ID Sex Age distance travelled (µm) Speed (µm/millisec)
C1 M 64 xxx xx
C2 F 48 xxx xx
C3 F 53 xxx xx
C4 M 51 xxx xx
C5 F 49 xxx xx
C6 M 57 xxx xx
C7 M 71 xxx xx
C8 M 63 xxx xx
Group 2 (Disease)
D1 M 61 xxx xx
D2 F 66 xxx xx ( a mutation found in this cell line only)
D3 M 63 xxx xx
D4 F 67 xxx xx
D5 M 51 xxx xx
D6 F 72 xxx xx
D7 M 64 xxx xx
D8 M 61 xxx xx
Hmmm...I would similarly ask Emily's question about clarification. I track several points:
1) Experimental Methods vs. Data Structure: your data illustration shows 1 row per cell line, but your methods states 10 replicates of each cell line with 300+ vesicles each.
- This would point to some variety of mixed/hierarchical model that can appropriately keep track of the nesting. As shown, you would be doing some sort of aggregation at 2 levels, almost like you're trying to force the data into a simpler 2 group comparison (e.g. t-test). That MIGHT give you the right answer...but I'd be very interested in how it compared to a more comprehensive model.
2) Distribution of dependent variables: How was distance travelled measured? Is it a "how far from start at the end?" ...or was motion tracked over time so that changes of direction all count as distance even if they're in the opposite direction. Either way, I suspect you can't have negative distances and so you stand a reasonable chance of having positively skewed data...potentially following a relatively simple distribution (e.g. lognormal) that could/should be modeled using a generalized model.
3) Theoretical: what are your hypotheses? From what theoretical model were they deduced? HERE is where you plan on how to deal with the potential for age and sex effects. Are they "nuisance" factors, merely adding their own systematic variability that can be effectively covaried out to increase power? ...or do they represent potential confounders to an inference claimin "disease" as the relevant distinction? ...or do they stand the chance to MODERATE the effect of disease? (e.g. difference between disease and control males, but not females)
Note that these are theoretical distinctions that lead to different hypothesis tests...each of which are equally rational with regards to statistical analysis.
It's the humans who have to work through these issues.
At the same time, you have VERY limited flexibility with regards to the number of cell lines, which is the appropriate unit of analysis unless there's something else going on in there.
Lastly, and this is related to your dependent variable, too...
4) with so many cells and vesicles...and the complexity of biology...I wouldn't be surprised if you saw multi-modal distributions...For example, what if cells in one part of the cell cycle had different vesicle speeds/distances than cells in other parts of the cell cycle. Under these circumstances, you may need to step into some variety of mixture model. Otherwise, you may end up averaging across high-speed and low-speed vesicles...and being let with a medium-speed central tendency (which few/no vesicles display) and huge variability.
Hi,
Agree with Jason about the need to consider clustering. Unsure about the advisability of using logistic regression with dependent variables measured (at least initially) on interval scales. Consider repeat measures analysis of variance to handle the correlation within cell lines with the repeat measures being conducted within each cell line.
Be well,
Howard
- Badges
- Science topic
- Similar topics
- Biological Science
- Cell Biology