...in mice...
For mice results in the top journals we're at about 1/3 applying to humans, which includes "has the effect and also starts killing you" if I remember correctly. And that's with the enormous selection bias of only counting the stuff in Cell and Nature and similar.
“ Regarding experiments on mice, some researchers have complained that "years and billions of dollars have been wasted following false leads" as a result of a preoccupation with the use of these animals in studies”