Predicting Human Facial Features From DNA Sequence

Alright, with the emotional rollercoaster of submitting my PhD thesis for defense over, and with my second dose of caffeine hitting my bloodstream, it's time to talk some science.

Just this morning, I came across a gem of a paper, published two days ago, that may just revolutionize forensic sciences and paleontology/anthropology (and it's open access! Bonus!).  This paper discusses the creation of a 3D model of human faces based only on their genome sequences.  I'm going to be talking about this paper in a forensics context: even though we've sequenced the human genome, generally only a few physical characteristics can be inferred from any DNA left behind at a crime scene.  Before 2012, forensic scientists could derive ethnicity, ancestry, and in some cases hair colour, but that was pretty much it.  In 2012, a Dutch group of researchers reported finding five candidate genes that influenced human face shape, or facial morphology.

Human face and head shape (or the technical term craniofacial shape) is determined during embryonic development through a series of precisely-timed gene expressions and interactions.  Then, as humans grow and develop, environmental factors and hormones continue to affect facial development.  Part of the reason why identifying the genes involved in human facial morphology has been so difficult is because through the use of genome-wide association studies, we've been trying to define facial development as univariate (meaning it only has one set of variables that define it), rather than the mutlivariate trait that it really is.  But this American group, led by Peter Claes, used a novel method that combined a bunch of different 3D modeling and gene analysis procedures, to look at between-population facial variation using participants with mixed West African and European ancestry from North America, Brazil, and Cape Verde.

The researchers used SNP (pronounced snip) genotyping to identify variance in the gene sequence of participants. SNPs are single nucleotide polymorphisms, which means that they are variations in the sequence of a gene at a single point in the DNA region being observed.

This may seem like it's just a tiny, insignificant change, but SNPs are actually responsible for the majority of genetic variation between humans, and are hugely important in the development of disease, and in response to pathogens, vaccines, and drugs.  In a human genome, there are many many SNPs, no one has just one.  They are also important in crop and livestock breeding, since they introduce the genetic variation that farmers and scientists select for (or against).  In my lab, I tried to work out a really interesting way to use SNPs to identify gene copy number, as an alternative to using Southern blots.  But that's a story for another day (unless it never gets published).

Anyway, Peter Claes and his colleagues selected 50 genes that were potentially related to craniofacial development, and a set of SNPs that had a high frequency of variation in these genes, to test these for their associations in facial shape variation.


Their method jointly modeled ancestry, sex, and genetic makeup (the technical term is genotype), in order to identify their independent effects on facial development. Once they created their 3D model they had another set of participants come in and look at the faces, and try and discern important characteristics from them.  Ultimately, the researchers found a set of 20 genes that significantly influence facial features, particularly in terms of face length and width, strength of the brow ridge, eye distance, nose width, and philtrum size and shape.  Obviously this is just a start, and there needs to be more study with other populations before this model is generalizable.  But the authors are confident that in 5-10 years, this model will be helpful in the prediction of human faces, especially the effect of other factors (age, temperament, adiposity) on facial features.  The construction of a 3D model of a person's face from DNA left at a crime scene could be used to help identify a suspect - although once apprehended, a sample of their DNA would have to match that found at the crime scene.  At the moment, this technology is too new to be used as evidence in a criminal trial.  Interestingly, this technique can also be used to make detailed facial reconstructions of our ancestors, which hasn't been possible to-date.