École Nationale Supérieure de
Techniques Avancées

Résumé

Heritability is a commonly used and important term to describe the proportion of phenotypic variability that is due to genetic factors. The accurate estimation of phenotypes’ heritability is a computational challenge for understanding the biological causes of disease that has been studied for decades in psychology and genetics but requires a vast amount of patients and as- certained phenotypes. Classical estimates of heritability are produced by high-dimensional pedigree-based tools, which can be time-costly and require high-end hardware components to be efficient. Plain and simple correlation model analysis is an alternative tool to compute heritability estimates efficiently from familial relationships reported in EHRs. Here we present a fast and accurate statistical method based on twin studies and various mappings of kinship relationships with non-parametric sampling from EHR data. We produced heritability estimates for dichotomous traits with several orders of magnitude less computational time than one existing and classic method based on the SOLAR software. These estimates were consistent with existing methods’ estimates and references whereas inconsistencies with literature were indicative of limitations to the simplified model as well as to EHR research. We conclude that this fast yet simple statistical method based on EHRs can provide quick and accurate insights on heritability estimates, for genetics and disease research.