search

Karrrramba / HIPrAM

Hydrophobic Interaction Chromatography Protein Profiling with Bayesian Additive Models
0 stars 0 forks source link

GAM functions #4

Open Karrrramba opened 1 month ago

Karrrramba commented 1 month ago

Optimize wiggliness through lambda so that the residuals of the alternative model equal zero. This would eliminate the instances where RSS of the null model < RSS alt model. (?)

Karrrramba commented 1 month ago

Does the distribution of the response need to be estimated and adjusted in the gam parameters first?

Karrrramba commented 1 month ago

One way to imporve the fit is to change the type of basis function in the smooth term via the "bs" argument. start with adaptive splives (ad) since for most profiles there will be lots of 0 intensity values, and try cubic splines (cr) or default thin plate splines (tp). Cubic regression splines are not as computationally heavy with large data sets. This might be easier then trying to determine the distribution of the response (comment #1).