Genetic Architecture of 3D Facial Shape in Diverse Populations

Predicting organismal phenotypes (characteristics or traits such as eye color and facial morphology) from genotype (DNA) data has important applications in human health, forensics, and biology. Within this context, this project is situated in the domain of imaging genetics, which refers to the use of anatomical or physiological imaging technologies as phenotypic assays in the context of genetic variations. The main objective of this project is to deploy machine learning, artificial intelligence, and image analysis in combination with statistical genetics to learn and deal with unknown interactions in predicting faces from DNA. In a first subtask facial heritability will be estimated using 3D facial images and genomic information from unrelated individuals. In a second subtask, knowledge of facial heritability will be harvest to improve the power of a genome wide association study on facial morphology. In a third subtask, the problem of predicting a phenotype from genotype is reframed as its inverse: Can we go from phenotype to genotype? This inversion, which we refer to as phenomic-prediction, might seem trivial at first, but allows the measurement of how well a given phenotype fits a target genotype, using powerful machine learning techniques. In a final subtask, testing phenotypes against target DNA also opens up the possibility of performing an actual phenotype prediction from genotype using artificial intelligence optimization techniques that simply evaluate and recombine phenotypes as solutions in an evolutionary algorithm.