[R-sig-ME] Analyzing evolution of resistance

[Evan Palmer-Young]

>
>
> Dear List,
> I'd appreciate suggestions on how to model a series of experiments on the
> evolution of resistance by a bee pathogen to nectar chemicals.
> I conducted an experiment in which I exposed pathogens to a constant drug
> concentration over 6 weeks. At each week, I tested drug resistance in the
> exposed lines, and also in control lines that were propagated in the
> absence of the drug. I have three replicate sub-cultured lines within each
> treatment (n=3).
>
> Here's a link to the DATA
> <https://drive.google.com/file/d/0B1YU9mSCw_jSbnpXVkYyWHJoTG8/view?usp=sharing>
> and my SCRIPT
> <https://drive.google.com/open?id=0B1YU9mSCw_jSb0MzZzZvMEtVYmM>
>
> I was planning to analyze this with a mixed model (lmer) of:
> *Drug EC50 ~ Treatment*Time + (1|Line)*
> to account for repeated measures on the lines in each week.
>
> However, the effect of time on EC50 is not at all linear.
> So my next idea was to standardize each week's EC50 values relative to the
> average EC50 in the control lines in the same week. So the model looks like:
> *standardized ec50 ~ Treatment*Time + (1|Line)*
> However, this seems to cause problems with the likelihood ratio test for
> the effect of time, because there is no variation in the control group: the
> standardized ec50 is pinned at 1.00 for each week.
> With this model, I get a df=0 when I compare full and reduced models, and
> p-values that are either 0 or 1. I feel like something is being divided by
> zero.
> I can still do chi-squared tests, but I preferred to use likelihood ratio
> tests, and I'm a little worried about the model because the likelihood
> ratio tests aren't behaving normally.
>
> My third idea was just to abandon the mixed model framework in favor of
> Tukey-corrected
> *t-tests of exposed vs. control lines ... at each timepoint*
> This would avoid the problem of explicitly testing the effect of time, and
> also avoid pretending that there was a linear effect of time.
> However, I'm concerned about just having n=3 at each timepoint, and also
> not having an overarching model for the whole analysis.
>
> What would you advice?
> Analysis on absolute ec50 values or ratios?
> Mixed-effects model or t-tests?
>
> Thanks very much for your suggestions,
> Evan
>
> Here is also a draft of the figure depicting changes in ratios over time,
> both for the (1) EC50 values (inhibitory concentrations of the chemical in
> the weekly "challenges" across a range of concentrations) and
> (2) Cell density ratios-- these show the degree of inhibition at a given
> concentration, to which the cells were constantly exposed​
> https://drive.google.com/open?id=0B1YU9mSCw_jSUGtJUWh3dkJPZHM
>
>
> --
> Department of Biology
> 221 Morrill Science Center
> 611 North Pleasant St
> Amherst MA 01003
> https://sites.google.com/a/cornell.edu/evan-palmer-young/
>
>


-- 
Department of Biology
221 Morrill Science Center
611 North Pleasant St
Amherst MA 01003
https://sites.google.com/a/cornell.edu/evan-palmer-young/

	[[alternative HTML version deleted]]