Summary of H2O2-LOH fitting, week of 2013 Dec 3-6

=>Dec4 I added fitting of half-black colonies, see  _2c1.2013Dec4halfblack.batch.logistical.R
Comparing the auto fit and manual fit show that I need to run gnls with weights.

=>Dec5-6, I worked on gnls with weights, _2a2.2013Dec4.stepBystep.gnslWeights.R
(This file was changed to stepByStep, because the automated fitting is too tricky to handle).

To mitigate bias of low colony counts, zero black colonies were adjusted to 0.5, because the lower bound zero is too stringent, and the upper-bound 1 is not observed. A mid-point point between 0 and 1 can be considered as the expectation.

I tried to work on a generic batch fitting code, but the initial values are so hard to automated, I decided to choose the supervised fitting (manual fitting) process, and fit every experiment one by one. I can manually input the fitting results into a summary excel sheet in file, _merged.tb.20131204_3pm.csv

There are two ways to fit the model. One is to average each H2O2 concentration first. The other is fit directly to the raw data. The first approach is much more numerically stable. The second approach has better theoretic support but often run into run-time problems. The second approach some times give better fitting, but can also give totally unreasonable fitting. After spending long hours on the second approach, it seems that I have to fall back to the numerically more stable first approach.

It is apparent that some viability ~ H2O2 curves fit better with exp(-x*lambda), the exponential survival function. For example, M1-2.

In M2-8, 06062011, dilution bias can be seen in the plot.

* It seems that viability - H2O2 look like exponential in normal plot, but sigmoidal in s~log(H2O2) plot. What does this mean?

Sometimes, gnls can really burn cpu, which suggest initial values are too far away. This can happen during manual fitting process when I switch from one strain to another and their responses to H2O2 are drastically different.

20131206, 11:41: I noticed the log(s)~log(H2O2) linear curve in sod2, file sod2,20120918,H2O2LOH.csv .  I need to do log-log plots in other strains to verify whether this is indeed due to sod2Delta effect.  I realized that many more experiments are need to show this is true. I am not what this means. So, I will not discuss this in the current manuscript. However, this is sth that is potentially interesting.

20131206, 13:17. First round of fitting is done, results in file: _merged.tb.20131206_13,16pm.xlsx
There are some missing Cb0.5 due to running bugs.

20131206, 14:50. After calculating the skipped one, the final results are in file:
_merged.tb.20131206_14,49pm.xlsx.