85 - Population Ancestry Influences the Accuracy of Facial Biomarkers to Diagnose Genetic and Rare Disorders

Poster Board Number: 85
There are separate poster presentation times for odd and even posters.
Odd poster #s – first hour
Even poster #s – second hour

Abstract Body : Introduction and Objective: Facial morphology is affected in 30-40% of existing syndromic conditions, and facial traits are commonly used for diagnosis of genetic and rare diseases (RD). Qualitative assessments and automated deep learning algorithms based on facial 2D pictures can be used to estimate this first diagnosis, which can guide further genetic testing. However, the facial gestalts of most RD are based on populations of European descent, disregarding the influence of population ancestry on facial phenotypes. Here, we tested the hypothesis whether the diagnostic accuracy of automated methods is affected by the admixed ancestry of a Latin-American population from the Southwest of Colombia, by quantitatively assessing the facial phenotypes associated to four genetic and RD.

Materials and Methods: We recorded the coordinates of 18 facial landmarks in 2D images from 79 controls 51 pediatric individuals diagnosed with Down (DS), Morquio (MS), Noonan (NS) and Neurofibromatosis type 1 (NF) syndromes, and quantified the facial differences using Euclidean Distance Matrix Analysis (EDMA). For each patient, we assessed the diagnostic accuracy of Face2gene, an automatic deep learning algorithm with widespread use in the clinical practice.

Results: Quantitative comparisons indicated significant facial dysmorphologies in DS (58.2%; P-value=0), MS (65.4%; P-value=0.013), and NS (47.7%; P-value=0.033). NF was associated with 11.4% of significantly different facial traits, but this pattern was not significantly different from random differences within control individuals (P-value=0.32) When clinical diagnosis based on genetic testing was used to verify the diagnosis based on 2D facial pictures, our results showed that Face2Gene top-5 accuracy was very high in DS (100%), moderate in NS (77.8%) and NF1 (66.6%), but low in MS (45.5%), with low gestalt similarity scores in highly admixed individuals.

Conclusion and Significance: Syndromes present characteristic patterns of facial dysmorphology that can provide effective early and low-cost biomarkers for genetic and RD diagnosis. However, to ensure that such diagnostic algorithms are broadly applicable to populations from around the world, the predictive models must account for population ancestry and incorporate populations with diverse contributions of Amerindian, African and European ancestry components.