Calculate marginal model posterior probabilities for one disease incorporating information from others
marginalone(STR, ABF, PP, pr, kappa, p0, fthr = 0.99, N0, ND)
| STR | list of models for diseases 1, 2, ..., n, each given in
the form of a character vector, with entries
|
|---|---|
| ABF | list of log(ABF) vectors for diseases 1, 2, ... |
| PP | list of posterior probability vectors for diseases 1, 2, ... |
| pr | list of prior probabilities for the models in STR |
| kappa | single value or vector of values to consider for the sharing scale parameter. the value of kappa=1 must be included, and if not will be prepended. |
| p0 | prior probability of the null model |
| fthr | models for all but the first disease are retained if
their cumsum(PP) < fthr. Ie set |
| N0 | number of shared controls |
| ND | list of number of cases for a set of diseases |
list of: - single.pp: list of pp for each model in
STR[[i]] for disease i - shared.pp: list of pp for each model
in STR[[i]] for disease i, - STR: not quite as input,
reordered so null model is first row - ABF: not quite as
input, repordered so null model is first row - kappa: as
supplied
Given a list of model matrices and log ABFs, this function calculates the marginal model posterior probabilities for thr FIRST disease without ever calculating the joint Bayes Factors for all cross-disease model configurations, which would require large amounts of memory.
This aims to be faster than marginalpp by ignoring contributions from other diseases to models with individual PP < eps. This is an approximation, and the effects of having too large an eps may induce inaccuracies. However, if you want to pull in information from several diseases, this may be the only way to do it within achievable computer time.