Generate a Better Model for Longitudinal Weight Gain

[davebridges]

The current model just uses fasting induced weight loss (pre-diet) to predict percent weight gain (via diet). We now have several other potential factors we could include to predict outcome, such as pre-diet hormone levels, weight, body composition, RER, food intake and VO2 (the latter three for both day/night cycles). Does adding these into the model improve the predictive ability?

[davebridges]

I created a new branch called cohort-data-entry in which i played around a bit with the cohort 9 code (see commit 8058b52). The changes i made were in the file scripts/HFD_eval_data.Rmd, i didnt play with cohort 8 or any of the modeling stuff:

• Fixed the data import, and wrote the raw data for that cohort into a csv file and commented out the import code (because it shouldnt be changing unless we find data entry errors).
• Eliminated CLAMS values with Volume.O2 < 100 or Volume.CO2 < 100. This should remove any measurements where the lid was open, or the flow was clogged.
• I generated an average for each mouse (for day and night) and then plotted that against percent weight gain. Those values, plus the MRI data should be good covariates for your modeling algorithm.
• I removed any measurements for CLAMS that were during the fasting phase, since that was mainly in order to figure out what is happening in the fasting phase, but probably isnt useful predictively.

If you think those help, then you should be able to merge in that branch to your main branch.

[qtran1]

Thanks! I'll take a look and incorporate this in my analysis if that helps.