Multiple non linear equations with shared parameters in brms

Modeling
1

I am trying to perform a multivariate analysis with three responses variables where each response variable has its own formula. But the three formulas shared parameters. I want to perform a Bayesian inference that results in one posterior distribution for each parameter. However, using brms I am obtaining one posterior distribution for each parameter for each response variable. How can I perform the bayesian inference obtaining just one posterior distribution?

I tried with:

for_tot<-bf(Vo~((VmaxGK4K6GTP^2)+(VmaxdGK2K4GTPdGTP))/((K1K2K4K6)+(K2K4K6GTP)+(K4K6GTP^2)+(K2K4K5dGTP)+(K2K5dGTP^2)+(K2K4GTPdGTP)),
VmaxG+VmaxdG+K1+K2+K4+K5+K6~1,
nl=TRUE)
for_cyc<-bf(Vo_cyc~((VmaxGK4K6GTP^2))/((K1K2K4K6)+(K2K4K6GTP)+(K4K6GTP^2)+(K2K4K5dGTP)+(K2K5dGTP^2)+(K2K4GTPdGTP)),
VmaxG+K1+K2+K4+K5+K6~1,
nl=TRUE)
for_lin<-bf(Vo_lin~((VmaxdGK2K4GTPdGTP))/((K1K2K4K6)+(K2K4K6GTP)+(K4K6GTP^2)+(K2K4K5dGTP)+(K2K5dGTP^2)+(K2K4GTPdGTP)),
VmaxdG+K1+K2+K4+K5+K6~1,
nl=TRUE)
form_two<-mvbf(for_cyc,for_lin,for_tot)

But this results in three distributions by parameters.

Thank you very much in advance for your help,

2

This is unfortunately not yet possible but will tentatively be possible at some point this year. The reason for it not yet being possible is that the required changes will break backwards compatibility in a relevant way and so will have to go into brms 3.0.

3

Is it possible to perform this using Rstan?

Thank you for your answer,

Santiago

1 Like
4

Which GH issue in brms 3.0 Milestone · GitHub addresses this? Thanks @paul.buerkner

5

It will be a natural result of internal refactorings of the brms model representation via sets of formulas.

1 Like
6

Is this possible now?
I have a similar use case.

7

Hello, is this still impossible with brms?