search

USEPA / CompTox-ToxCast-tcpl

US EPA's Toxicity Forecaster (ToxCast) Pipeline. More information on the ToxCast program available here: https://www.epa.gov/comptox-tools/toxicity-forecasting-toxcast
https://cran.r-project.org/package=tcpl
Other
29 stars 12 forks source link

invitrodb.mc5 column "err" #237

Open kpaulfriedman opened 5 months ago

kpaulfriedman commented 5 months ago

Mc5 column “er” has negative numbers. Is this the log10(error), where error = sigma^2? I thought this might be a result of the strangeness of moving from fitting on log10 scale vs fitting on arithmetic scale? We should investigate the code to see what is computed and stored and consider if it should be changed to arithmetic units.

kpaulfriedman commented 4 months ago

@sedavid01 and @brown-jason I had a conversation today that made me more confused about what "er" is in mc4. It is not clear to me if this is natural log of sigma, or is it transformed to some variance or SD measure? There is some inconsistency in the tcplfit2_core vs tcplhit2_core functions in tcplfit2. We looked at tcplfit2_core.R in tcplfit2 and v3_schema_functions.R in tcpl and could not figure out mathematically how to define the "er" term...somewhere it must be renamed as "er" from "err" as well and I couldn't quite find it....thank you!

brown-jason commented 4 months ago

@kpaulfriedman er - error term- is a parameter that is defined in all of the models' fitting in tcplfit2. tcplfit2_core calls each model fit and er is a term in the constrained optimization that occurs during fitting see fithill.R for example. We also store the standard deviation of all parameters with the _sd suffix. I don't follow the er vs. err renaming because everywhere I see the term it is defined as "er". @sedavid01 can provide more context on the mathematical meaning of the 'er' term.