Scientists Use Genome Sequencing and Machine Learning to Predict Facial Traits

Some of the recent advancements in technology and science are downright scary. A recent experiment by researchers at Human Longevity Inc shows how their software is capable of rendering faces based on DNA samples. It is an interesting development in forensic science, though one that is disconcerting in many ways.

Facial Trait Prediction Through DNA Samples

It is pretty cool to know scientists can effectively predict physical traits based on our DNA. To do so, they used whole genome sequencing data and machine learning. The findings of Human Longevity researchers have been published in a well-known scientific journal. So far, the study is getting some good feedback, although the general public may be rather concerned about this development.  If scientists can “generate” one’s face from a DNA strain, what else could they do?

The study was designed to point out how forensic science can make use of newer technologies. Right now, DNA samples are used to find a match in the system, which is a pretty powerful solution. However, by using machine learning and whole genome sequencing data, this concept can be taken to a whole new level. There is also a data privacy issue. Without informed consent, use of such technologies may be considered to be controversial or even illegal in some jurisdictions.

The current trend involves placing more genomes in public databases. That is both obviously good and bad. In this particular study, researchers took samples from 1,061 ethnically diverse people. The team also collected data from 3D facial images, voice samples, and so on. Out of all the data collected, predictions of eye color, skin color, and sex were highly accurate. More complex genetic traits were found to be a lot harder to accurately predict.

Although the predictive models developed by the researchers are seemingly sound, there is still a lot of room for improvement. The machine learning algorithm utilized in this test found combinations of all predictive models. Approximately eight in ten participants were successfully identified. African-American and European participants, on the other hand, saw a success rate of “just” 50%. That was not the result the researchers had hoped for, though things have progressed in the right direction.

Human Longevity Inc’s co-founder Craig Venter states:

“We set out to do this study to prove that your genome codes for everything that makes you, you. This is clearly a proof of concept with a limited cohort but we believe that as we increase the numbers of people in this study and in the HLI database to hundreds of thousands we will be able to accurately predict all that can be predicted from individuals’ genomes.”

It is good to see the team acknowledge the privacy risks associated with this research. There need to be better safeguards and policies for individual privacy in the genomics era.  It is certainly true that imaging technologies, combined with machine learning, can yield some unexpected results. Whether or not we will see more of these developments in the coming years remains to be seen.