Assignments and Discussion Issues #2

[rainersachs]

Starting a continuation issue because the first one is very long

[rainersachs]

Here is a preliminary version of part of the web supplement that we will put on the Journal's web site. It will act as a paper appendix with almost unlimited space. This preliminary version touches on many issues that need to be discussed when preparing the master copy of the main script we will need , first for the minor paper and eventually for the modeling part of papers reporting thew results of recent HG gland experiments.
ESM_Online_Resource_1_no_endnote_magic.docx

[rainersachs]

The main papers on this HG data are the following; you probably have them. They include 2 papers GitHub refuses to load:
Alpen EL, P Powers-Risius, SB Curtis and R DeGuzman. "Tumorigenic potential of high-Z, high-LET charged-particle radiations." Radiat Res 136(3): 382-391. (1993).
And Alpen EL, P Powers-Risius, SB Curtis, R DeGuzman and RJ Fry. "Fluence-based relative biological effectiveness for charged particle carcinogenesis in mouse Harderian gland." Adv Space Res 14(10): 573-581. (1994).

They also include Chang 2016, and a 2017 Cucinotta paper
17CucNTEhiRiskME.pdf
16EllieCucPollyHG.pdf

They also include the previous URAP paper and the Web supplement to that paper: Siranart N, EA Blakely, A Cheng, N Handa and RK Sachs. "Mixed Beam Murine Harderian Gland Tumorigenesis: Predicted Dose-Effect Relationships if neither Synergism nor Antagonism Occurs." Radiat Res 186(6): 577-591. (2016).
16Synergy12.27.16RR.pdf

Finally, here is a first draft of part of the upcoming minor paper. It contains various items that need discussion.
MS05.docx

[rainersachs]

Hi Edward: In our discussion yesterday you asked about a 4-panel Figure. The figure is shown as a .png in MS05.docx I downloaded above as Fig. 2.2.4.1. I have an. eps version but GitHub won't accept that. The Fig. must be from a ggplot2 in the master and should be easy to resurrect. I suggest we ignore this issue for now and if necessary resurrect Fig. 2.2.4.1 later when we have the draft of the minor paper in better shape.

[rainersachs]

We have to add vertical error bars to some of the graphs when using plot( ). Googling r: error bars gave me a number of hits, including
ftp://ftp.uni-bayreuth.de/pub/math/statlib/S/Harrell/help/Hmisc/html/errbar.html.
which talks about the function errbar( ) in package Hmisc that gives exactly what we need. But I cannot find any package Hmisc nor any function errbar( ) using my RStudio sandbox no matter where I look.
Edward: can you find them?
Perhaps this can be done only within ggplot2( )?

[eghuang]

Try running install.packages("Hmisc") to download the package.
Then library(Hmisc) should load the package in RStudio and you should be able to use its functions normally.

[rainersachs]

Thanks. errbar( ) works nicely.

[rainersachs]

Here is a file which needs to be checked for consistency with the raw_data_ordered.csv file
hg_raw_data_4.4.18.xlsx

[rainersachs]

a version of the paper and a figure proportions file; see comments to the changes in plot.R
MS07.pdf

[rainersachs]

Here is a comment on major-paper and programming plans for fall semester 2018.
URAP CA and mouse pods.pdf

[rainersachs]

URAP CA and mouse pods.pdf

The figure above and in the .pdf above had a mistake. Here are corrected versions of both

[rainersachs]

Hi Edward:
A coordinated set of 7 files , rks_something, I just uploaded works together well, except for the following.

My PI asked me for a ribbon plot of the mixture defined in rks_plots.R starting on line 215. If I can get them to her ASAP good so for now this becomes top priority. If not, no problem. I can put her off for a while.

But trying to program a plot for that mixture I ran into what I think are probably two or more (comparatively minor) conceptual mistakes in rks_monteCarlo.R and/or rks_synergyTheory.R which probably need to be fixed for stability.

The first symptom is that uniroot( ) sometimes can't find the root it is looking for and for the line 215 mixture in rks_plots.R that is limiting me to absurdly small doses and/or absurdly small numbers (n=10 instead of n=500) of monteCarlo runs. The second symptom is that when I do limit like that I can get an answer but find that IEA baseline, instead of lying between the mixture component DERs lies in the center of the ribbon. The ribbon itself is some absurdly high, narrow shape, which is not surprising since it is based on n=10.

to deal with the second symptom first, I think probably I(d) is being calculated as the midline of the ribbon rather than by its differential equation. That would be OK if we were using something like n=10^8 but for practical values like values like n=200 or n=500 it is needlessly very approximate. The differential equation is much more accurate and more direct since the Monte Carlo has to solve differential equations in any case.

The uniroot problem has occurred before and you cured it by making the search interval much larger. That should in fact always work here because all our DERs, barring unlikely negative values of the adjustable parameters, are monotonic increasing and approach 1 for large doses. But it is also rather ugly. A better and more reliable fix should be using "upX" or "downX" in our monotonic increasing case.

Unless I'm being overoptimistic these are not killer problems, but I do think that in the long run they should be fixed. And there is no strong reason to hurry so don't feel any obligation to do anything before you get back here, or even then really, if your other projects are going well

[rainersachs]

Yimin: Sooner or Later the mouse pod will write a major paper jointly with the experimentalists. There may be a less major paper first. We have all the main elements of the needed scripts in fairly good shape. We just have to clean up many details. This academic year we need to clean up all those details. When that is done you can return to the more mathematical project you were working on at the end of last semester.

1. Please get familiar with the files labelled rks_foobar in the main branch here. They are the only ones we will use and improve. They implement a flow where all data entry is via .csv files; .R files never change the data, they just use it. The flow chart is as shown here.
   

2. One minor bug that can be and should be fixed now in these files is that sometimes data base entries are accessed by numbers. They should instead always be accessed by column names, e.g. db[ion=He,] not db[5:12, ].

3. A more messy detail is that we have to add some of Cucinotta's models in Chang et al. 2016 as extra models and compare them to our preferred models. Please read the modeling part of Chang et al. (you can find the reference in the rks_data. module; its name is something like rks_dataAndInformation.R). Come to our Thursday Aug 23 meeting (1:`10 my office) prepared to ask lots of questions about the models in the Chang paper. We will need to chase down input values for what Cucinotta calls Z_eff^2/beta*^2 in his modeling section of the Chang paper.

4. In general please always prepare for our meetings by looking here at this issue in GitHub.

5. I have now added a couple of commented lines to the end of rks_DataAndInformation.R which show how to calculate Z_eff^2/beta*^2 and then ion speed relative to the speed of light. We will do those calculations somewhere, e.g. in .xlsx, add them to the input .csv files, and use them in rks_synergyTheory.R

[rainersachs]

16EllieCucPollyHG.pdf
Yimin: in addition to your studying and tweaking the new file structure (see comment from 3 days ago) and studying the very much speeded up Monte Carlo method in those files, the other main part is adding what is called the NTE1 model in the .pdf uploaded here of the 2016 paper by Chang et al. referred to in the rks_DataAndInfo script. Please start by reading everything about the "NTE1" model in the subsection "Statistics and Modeling" which begins on the bottom of the first column on p. 451. Skip Eq. 1; skip everything about the TE and NTE2 models, neither of which we will need. Read everything about NTE1 up through Eq. 6. You will find it very hard, so you will need to read many times and make notes on many confusing points to ask me about.
Skip the rest of the subsection. Please try to finish these assignments within a month. Then there will be one more hard part (Reading some of the results section and implementing that). Thereafter things should be much easier and more fun.

[rainersachs]

18fa_Cucinotta_NTE_for_private_astronauts.pdf
Here is a very up to date paper that deals with almost all of the modeling issues in our calculations, uses almost all the acronyms with which most of you are by now familiar, and also gives the space-travel background for the modeling. I suggest you read the article as deep background for the projects. Among the terms that we haven't discussed or used much are Relative Biological Effectiveness "RBE" and "Quality Factor". They are not rigorously defined but even as fuzzy concepts they are quite important, so you might want to look them up.

[rainersachs]

LET:
(https://github.com/sachsURAP/NASAmouseHG/files/2321242/Ballarini_2008_New_J._Phys._10_075008.pdf)
Elementary_picture_of_LET.pdf
Here is a .pdf, Elementary_picture_of_LET, that I wrote because I noticed that I had never properly explained LET, which we use so much, even in naive terms. The Ballarini paper, also given above, shows some figures that illustrate a more sophisticated model than the one I used: Figure 1 lower right corner and figure 2.

[rainersachs]

Reviewer #1 REBP
the reviewer asked a bunch of statistics questions. Dae Woong Ham , a coauthor on the paper knows the answers well. Since some of you don't know each other well l this comment is to introduce you to each other and also ask you to work out among yourselves how to handle the statistics questions reviewer 1 asked. I will now leave that entirely to you. Please decide who will make a report to me that enables me to answer the statistics questions in Review 1 by Oct. 30. If extra calculations or figures are needed, please make them. Yimin is interim in charge until one or more of you volunteer to take responsibility. Here is the relevant contact information of the coauthors.
edwardgh@berkeley.edu
daewoongham@berkeley.edu
(Claire) Yunzhi@berkeley.edu (Yunzhi Zhang)
yiminlin@berkeley.edu

[rainersachs]

Yimin:
Just to double check if we are working on the same page. Is the following consistent with what you are doing:

1. The NTE1 model in 2016Chang has 8 adjustable parameters:
   alpha(L) contains 3 alphas
   lambda(L) contains 3 lambda
   kappa(L) contains 2 kappas
   the values Cucinotta got for these is given in Table 6 of Chang et al.

2. These 8 parameters should be calibrated by regression using only dose points >0 and only for HZE (i.e. not protons or helium ions).

3. When you get values for these 8 please send them to me. Then I can check if I think their might be an error, using the values and 2016 Chang, without looking at your script. The values should not be the same as in 2016 Change for several reasons -- one reason is that you are probably using dose in cGy, a factor of 10^2 compared to using dose in Gy. So the parameters may be off by factors of !0^2 or 10^4. But I can correct for that and for other differences to see if your results are compatible with the numbers in 2016Chang. There should be some similarities.