Within the field of Forensic DNA Phenotyping, the accumulation of reliably predictable externally visible characteristics, from pigmentation traits to age, can aid investigative intelligence to help find unknown trace donors in serious crime cases that lack known suspects. Adding facial traits would be highly invaluable; however, predicting a face from DNA is challenging due to the large complexity of the phenotypic and the underlying genetic variation of the human face. In recent years, significant progress was made on both, the genetic side of genome-wide association studies (GWAS) with increased sample size and the phenotype side of mesh morphometrics with global-to-local and distance-based facial traits used in GWAS. With its prominent position in the facial midline, making it a key facial descriptor in eye-witness testimony, the nose may be deemed the first facial trait worth exploring for genetic prediction. 

Here, we tested all currently known face-associated DNA variants for their value to predict nasal morphology. A total of 2274 significantly associated DNA variants were compiled from all previous face GWASs, including the most recent and largest study using over 50,000 individuals, and were evaluated in a global set of 1617 individuals (995 Europeans and 622 non-Europeans) using a data-driven morpho-space created by a template mesh to digital 3D facial scans for generating dense quasi-landmark data of the nose. Meshmonk, an open-source toolbox was utilized in combination with Procrustes alignment and Principal Component Analysis to produce a low-dimensional morphological “nose-shape space”. Independently contributing DNA predictors were identified via a multi-step selection framework, with initial filtering based on permutation tests per-variant across all nasal features followed by partial least squares regression combined with variable importance metrics to account for trait correlations and to avoid genetic redundancy. Model parameters were optimized using 10-fold cross-validation. Using independent DNA predictors, prediction models were built and internally validated. External model validation using independent data was performed on 325 Europeans and 175 non-Europeans. Prediction outcomes were visualized by projecting them onto a global nose-shape morpho-space and compared to their original image to illustrate performance metrics.

This study represents an important step in demonstrating the feasibility of predicting facial features from DNA. More advances in genetic face prediction will largely depend on continuous progress of cataloguing face-associated DNA variants more comprehensively. Until the genetic face architecture is unveiled more completely, DNA-prediction of certain parts of the face is a more plausible goal as opposed to predicting the entire face.

Authors

• Manfred Kayser (Erasmus MC University Medical Center Rotterdam, Netherlands)
• Ziyi Xiong (Erasmus MC University Medical Center Rotterdam, Netherlands)
• Mallory Price (Indiana University Indianapolis, United States of America)
• Lauren Huntington (Indiana University Indianapolis, United States of America)
• Subhashree Ramanathan (Indiana University Indianapolis, United States of America)
• Susan Walsh (Indiana University Indianapolis, United States of America)
•  Kayla Borowski (Indiana University Indianapolis, United States of America)