search

NCI-CGR / plco-analysis

Primary workflow for the PLCO "Atlas" project
2 stars 3 forks source link

categorical trait module expansion #4

Open lightning-auriga opened 3 years ago

lightning-auriga commented 3 years ago

existing method is a hack with expected anticonservative bias. bakeoff different scalable regression methods

lightning-auriga commented 3 years ago

longtime hope for the project has been to add additional analysis modules under shared-makefiles (can easily be snakefiles or other languages). one such candidate would be a categorical trait method that doesn't rely on the N_categories - 1 min(p) Bonferroni correction currently used, which is biased towards smaller p-values due to the naive min operation.

lightning-auriga commented 3 years ago

current handling of min(p) method is in this associated software package

lightning-auriga commented 3 years ago

current accumulated requests from collaborators/supervisors below

lightning-auriga commented 3 years ago

Below is a reference and software that could address the question we had yesterday on analyses of categorical variables (in case you have not seen it):

manuscript: https://www.biorxiv.org/content/10.1101/2020.10.09.333146v1.full

software: https://github.com/WenjianBI/POLMM

Another suggestion from a collaborator is to incorporate customized (R) codes for proportional odds/polytomous regression in existing pipelines. EPACTS (https://genome.sph.umich.edu/wiki/EPACTS) and PLINK allows this kind of customization though the run time might be an issue.

lightning-auriga commented 3 years ago

(note that the above was originally based on a recommendation based on methods for subtype analysis)