3D modeling of face shape by human genes

Hi, Habr! We are a biotech startup "My Gene". We want to bring to Russia a brave new world of personal genetics. In our blog, we plan to post the latest news in this area, to present interesting scientific articles on bioinformatics, and probably even tell you one day what it is like to be bio-start in our country. We hope that the concepts of “cloud lab” and “the genetics of things” (puns intended) will soon be included in our lives, and we will contribute in every way with our efforts.



Today we will tell about a recent article in Plos Genetics, about a new approach to the genetic prediction of facial features. This case is especially interesting in contrast, if we recall the recent news on a similar topic, for example, about an artist from New York who is printing portraits on a 3d printer supposedly using DNA tests. It’s one thing - a buzzword window dressing, and another when real scientists do it.



Prior to this study, all attempts to find a connection between the shape of the face (yes, in principle, and any other difficult formalizable trait) and genetics used the simplest approaches to facial evaluation, since genetics traditionally used one-dimensional variables for association (for example, does a person have a specific disease, or how tall he is; see GWAS ). In the worst case, they simply measured the distances and angles between the eyes, the nose, and other features of the face, which may differ between or within different populations. In more advanced studies, the main component method was used, which, on the basis of such measurements, represented a number of abstracted metrics (such may be, for example, already “east”) in the form of, in fact, the main components.

')





The methodology of this study (the pictures are clickable, descriptions there) begins with a three-dimensional scanning of the subject, during which the face area is distinguished, the approximate direction of the face is determined by five main points, and an anthropometric mask is applied consisting of 7150 "pseudo landmarks" (quasi-landmarks, translation author; it’s not obvious from the text of the article whether they were calculated in the course of work or whether ready ones were used, most likely the last), and this is done in two approaches with averaging (more precisely, with it Procrustes superimposition ): first on a normal scan, then at his mirror image to get rid of the natural asymmetry of the human face. The obtained multidimensional data from 592 people - and the study involved citizens of the United States, Brazil and Cape Verde (Cape Verde Islands) - all were used in the same method of the main components, 44 components were identified that are responsible for 98% of the total variability. The impact of the 10 most significant components can be observed below:







Next comes hardcore statistics, namely, BRIM’s own method (Bootstrapped response-based imputation modeling), for implementation details it’s easier to immediately refer to the inconspicuous Supplementary text. The method is based on PLSR ( partial least squares regression ) and essentially produces a hybrid variable from independent and dependent regression variables, called the RIP (response-based imputed predictor). The latter can be used instead of independent variables and get a higher order RIP, this is the bootstrap. So, several different hybrid variables were obtained: RIP-S, expressing the dependence of a person on the floor; RIP-A, from genetic origin; and finally the RIP-G series, from specific genetic polymorphisms. The influence of RIP-A and RIP-S, expressed in the variability of the mask areas, the difference in their area and inclination, and the deviation along the normal, so to speak, is evident:







In addition, the article showed a good correlation between hybrid variables and the natural assessment of the same factors by people (that is, they conducted a banal opinion poll, “how much this person seems feminine to you, is it similar to Negro”), and discovered the effect of individual genes and polymorphisms on some facial features.



Obviously, such research has the greatest practical significance in forensic science. Already now (about seven years already), having a thousand or other DNA samples of people of different nationalities, we can draw up a genetic map and predict the nationality (sorry, meaning ethnic composition) of any criminal who left a piece of himself as evidence. Now it became possible not only to narrow the circle of suspects, but also to make a photofit. Paranoids and conspiracy theories, by the way, do not particularly worry about biometric passports - they use (if at all) the technologies of the last century, suitable only for comparing 1-to-1, as well as fingerprints work. Genome-wide analysis is still too expensive for any massive use.



However, you can go a little further and in a slightly different direction. Many have probably heard about the services of IVF (in vitro fertilization). Again, even now the most advanced clinics offer, in the appendage, also preimplantation genetic diagnostics, which allows choosing the embryo of the desired sex and obviously without some diseases. Now imagine that you can also draw the face of your child. And then throw +5 to stamina and +3 to intelligence. And then ... Who said "Gattaka"? There are fears that the mention of this film with each news about personal genetics will soon become a bad form. We will no longer, we still have to sell it.



Source KDPV: www.spiegel.de/fotostrecke/erbgut-bestimmt-die-form-von-gesichtern-fotostrecke-112776.html