In theory, genomes shared anonymously could be linked to people on social media because the DNA can be used to predict facial features, but the risk is vanishingly small
17 November 2021
The researchers studied the genomic data and online photos of 126 individuals, then tried to match faces to genomes. They worked backwards from the faces, using AI to analyse the photos and predict gene variants, then looking for genomes with those predicted variants.
Given a subset of just 10 individuals, the team was able to identify a quarter of them. However, as the number of people increased, accuracy plummeted. For groups larger than 100 people, it was negligible.
Venkatesaramani and his colleagues say a key reason for this is that social media images are much lower quality than the studio photographs used in previous studies.
Daniel Crouch at the University of Oxford, who has studied the genetics of facial features, agrees that the risk is low. But he says the team’s analysis shows that this is actually due to the difficulty of linking gene variants with specific facial features, rather than image quality.
“It is not really the quality of photos that matters that much,” says Crouch. “We are still only really just starting to understand the genetics of facial variation.”
“Once our understanding of facial genetics improves, our ability to link faces and DNA will improve too,” he says. “However, I suspect we will never quite get to a point where we can predict whether a DNA sample belongs to a specific person, drawn from anyone on the planet, at least in our lifetimes.”
The claim that there was a serious risk that people whose genomes were being used for medical research could be identified from photographs was made in a 2017 paper. This study was heavily criticised for containing major flaws, says Crouch.Journal reference: Science Advances, DOI: 10.1126/sciadv.abg3296