The comparator pool sets the candidate list in a 73-preparation dietary screen in high-fat-diet mice

Background Screens of dietary preparations test many candidates against one shared control group. The reference chosen for a candidate decides which candidates emerge. Methods In this study, we employed a resource of 824 male C57BL/6 mice. That resource contained 86 herb-coded preparations administered in the diet across eight cohorts. Of these, 73 met the pre-specified criterion for the comparative screen in six cohorts. The screen yielded 583 preparation-versus-pool comparisons in a metabolic family of 365 tests and a tissue family of 218 tests. We compared each preparation within its cohort with the preparations tested alongside it and with the untreated high-fat-diet control. We held the false discovery rate within families by Benjamini-Hochberg control. We also ran a step-down max-statistic permutation across the whole comparison set, with a model-validity precondition. Results Body-weight gain and fat-depot mass were the two endpoints whose high-fat-diet model was established in all eight cohorts. The two references agreed in 83.4% of comparisons, and their block-level mean difference was equivalent within +/-10 percentage points. The comparison-level differences lay between -33.3 and +31.4 percentage points in 95% of cases. We decomposed the spread: one percentage point of difference between the comparator pool and the untreated control added 0.99 percentage points of disagreement (R2 = 0.87). The 14-fold difference in the size of the two references added none (R2 = 0.00). Three preparations reached family-wise significance with step-down adjusted P values of 0.000-0.022, and each was selected in every one of 1 000 bootstrapped resamples. Seven reached q < 0.05, and among those seven body-weight gain was 27.4-70.6% lower in five and fat-depot mass was 42.5-69.0% lower in five. Nine reached the exploratory level. The candidate list kept a stable core and a specification-dependent periphery across 64 analysis specifications. Those specifications vary the reference, the unit of analysis, the multiplicity convention and the threshold. That core was DaSS 94%, LuHu 88%, YiRn 78% and JuFn 75%. Conclusion Reading the comparator as a method makes the candidate list a statement about which preparations survive which reference.