3D model reconstructs face from DNA sample

Scientists are getting closer to constructing a likeness of person's face using nothing but a DNA sample. Postdoctoral researcher Peter Claes and his colleagues describe the technique in a recent publication in PLOS Genetics. Their work opens a horizon of potential future applications in forensics, anthropology and medicine.

At left, a side-profile photograph of a volunteer. At right, the 3D model created from a sample of her DNA | © New Scientist

The days of police composites based on vague eyewitness descriptions are numbered. In the not-so-distant future, a strand of hair may be all that is needed to create a virtual ‘mugshot’ bearing a reasonable likeness to its owner. “I am enthusiastic but also cautious; enthusiastic about what is already possible, and cautious because we don’t want to encourage unrealistic expectations,” says Peter Claes (Processing Speech and Images Research Team, ESAT/PSI and Medical Imaging Research Center, University Hospitals Leuven). “We are not yet at the point where we can create a guaranteed accurate and complete prediction of a face on the basis of, for instance, a saliva sample. What we have demonstrated is that facial recognition on the basis of DNA is possible, and that we are moving in the right direction.”
The partnership that eventually led to the paper in PLOS Genetics began six years ago, with an email from Mark Shriver, professor at Penn State University. “Mark contacted me after seeing my doctoral dissertation, which was about visualisation techniques for craniofacial reconstruction – craniofacial reconstructive surgery. He had a large database of 3D images of faces and matching DNA data. The idea to visualise a face using genetic material grew from there.”
“It took a significant amount of work to bring our respective domains – medical imaging and genetics –together. I have the feeling that things have now kicked into high gear, although I continue to emphasise that there is still much work to be done.”

Rich description

“Our face says a lot about us. It provides information about our health, gender, emotions, and so on. So it is also very complex. And yet, scientist still describe the face using simplistic parameters: nose length, eye colour, distance between the corners of the mouth, etc. The facial features that can be predicted on the basis of genetic information alone are limited. This made it difficult to arrive at a good description or a reliable computer-generated representation. We wanted to change that.”

"The technique could eventually be used to compare crime scene DNA with faces recorded by nearby security cameras"

“We based our research on a rich genetic basis. We were able to designate twenty genes that influence the shape of the face. Of course, this is not a complete determination: other genes, our age, the environment, our habits and so on also influence the shape of the face. But this is, in any case, a good starting point.”
“We used a stereoscopic camera to capture 3D images of almost 600 volunteers from populations with mixed European and West African ancestry. Because people from Europe and Africa tend to have differently shaped faces, studying people with mixed ancestry increased the chances of finding genetic variants affecting facial structure. We fitted a grid with 7,000 points to the 3D images of our volunteers. This made it possible to arrive at a much more accurate description of facial shape than the nine points that were used previously.”
“We entered the DNA information and the facial measurement information into a computer, together with data on the gender and the genomic origin of the volunteers. This allows you to correlate a highly dimensional thing like a face with a large amount of small and interrelated information packets from the DNA, which, in turn, makes it possible to create a virtual reconstruction or prediction of the face. As our early findings show, the result comes very close to the real face.”

Neanderthal

Claes and his colleagues are now further refining the technique. “We’re looking for extra genes that play a role in determining a given facial feature. But we are also investigating the opposite track: what is the DNA of people with certain facial features? This could allow us to help the police narrow searches by comparing DNA found at a crime scene with the faces of thousands of passersby recorded by a nearby security camera, for instance. But the technique could also be applied to help diagnose certain diseases: what genetic problems do people with specific features have?”
The technology also has potential uses in anthropology: “We could use potentially use the technique to get an idea of how our distant ancestors looked. Using ‘fossil DNA’ that Europeans inherited and Asians did not, you can isolate genes that code for facial features, which could give us a realistic image of our Neanderthal ancestors.”
A bright future, with plenty of light still to be shed: “We are pleased with the results, but we realise that many years of fine-tuning and broadening are ahead of us before we can make these kinds of applications concrete.”
Ludo Meyvis and Jack McMartin