Collapsing models focused on PTVs explained 80% of binary and 55% of quantitative associations. Remaining signals emerged from models that included missense variants. While these results confirm the importance of PTVs, they also emphasize the role of other forms of variation in human disease. We found that using the missense tolerance ratio (MTR) to retain missense variants only in constrained genic sub-regions improved the signal-to-noise ratio. Specifically, 15% (133 of 878) of significant relationships detected via the three MTR-informed models were not detected in analogous models that did not incorporate MTR35. Moreover, for phenotype associations where both MTR and non-MTR versions of a model achieved significance, effect sizes were significantly higher in the MTR-informed versions (Mann–Whitney test P = 0.006; Supplementary Fig. 3). Thus, MTR appears to effectively prioritize putatively functional missense variation in collapsing analyses of complex disease.
