Focus: DNA-based Facial Portrait
ZHAO Wenting, JIANG Li, LIU Jing, ZHAO Lei, MA Xin, JI Anquan, LI Caixia
DNA identification plays an important role in criminal investigation. But when the evidential DNA does not match either a suspect’s STR profile or any other one’s in a criminal DNA database, and/or the mass DNA screening and family searching failed to give any information for tracing unknown forensic sample’s donors, cold cases would be even caused, leaving them to be waited for various periods of time (likely very long) to solve. Predicting human externally-visible characters (EVCs) of an unknown person is an emerging contrivance by which to provide investigative hints allowing to trace the suspect, who is not identifiable via conventional comparative DNA profiling. In recent years, the potential of constructing useful DNA-based facial portrait is of great interests in forensic studies. Facial morphology is a combination of many complex traits, highly heritable, because much of the total variation in facial features is genetically mediated. At the early-stage of the kind of researches, the knowledge of genetic variation on facial morphology has mainly arisen from studies of genetic abnormality such as non-syndromic cleft lip with/without cleft palate (NSCL/P), and from those of developmental mechanisms, including PAX3, FGF, GHR and many other signaling pathways. In these studies, facial characters were traditionally simplified by straight scalar measurements, e.g., nose-width or eye-distance, therefore having no differences from the measurements of height, BMI or pigmentations. However, facial morphology is such an extremely physical complexity that dimensionality reduction techniques, where each principal coordinate is treated as a scalar trait, can cause the statistical power decreased dramatically. Recent advances have produced novel methods for fully automatic 3D facial image mapping and crude models of facial structure constructed by SNPs variants analysis. The non-rigid registration method or spatially dense quasi-landmarker was applied for fully automatic 3D facial image mapping and aligned faces via high-density pixel points. Based on these methods and joint modeling of sex, ethnicity and genotype, the independent effects of particular alleles on facial features will be uncovered and a 3D facial prediction model could be built. In the future, such developments will lead to a more detailed description of an unknown person’s appearance from genetic variations, very likely to expedite police investigations by providing more scientific information of forensic evidence.