A-T-C-G to F-A-C-E: what are the odds you’d look like that?

Identity, parentage, innocence. These are some of the things that can be determined with even just a small amount of DNA. But what if… we can do more? What if I can infer what you look like without ever having met you, just by looking at your DNA? Scientists strive to do just that.

DNA – that seemingly tiny double-stranded thread composed of As, Ts, Cs, and Gs found in every single one of our cells. It is genetic material that acts like a blueprint – our eye colour, hair colour, skin colour, blood type, and other characteristics all planned and written in every single strand. Although the expression of these characteristics can change depending on the environment, DNA can still tell you a lot about a person. This is the premise of DNA profiling.

DNA technology and DNA profiling are widely used in numerous circumstances in our world today. Chances are, you’ve probably heard about DNA profiling before. 23&Me and other ancestry services use your DNA to determine your ancestral background. “You Are / Not the Father” type shows, where they perform tests to determine a baby’s paternity depends on DNA technologies that analyse if sequences from the baby’s DNA can also be found in the genome of the suspected father. Both of these are examples of DNA profiling, wherein they use DNA to determine a person’s identity. But perhaps the most widely known implementation of DNA profiling is with regards to forensics, largely thanks to the numerous police, detective, and investigation type shows prevalent in the media. Situations where unknown DNA was found in a hypothetical crime scene and must be compared with a suspect’s DNA in order to determine innocence or guilt, and if it’s not a match, then the unknown DNA must be run through a database of known DNA profiles in hopes to find a match and a suspect. But… what if there’s no match? A paper by Sero et al. published in 2019 proposes a method in the case that this happens.

A flowchart of the strategy. First, acquired DNA from the scene will be matched with DNA from a suspect. If it fails, the DNA will be run through a genetic database, containing the genetic profiles of known individuals. If that fails, the DNA will then be run through a phenotype database. Each face-to-DNA classifier within this database of known faces and identities will be matched and scored against the unidentified DNA sample. If that also fails, then facial characteristics inferred from the DNA sample can be used to predict the phenotype of the unidentified person to whom the DNA belongs.

While previous attempts have been made to infer someone’s face simply from their DNA, Sero and colleagues propose the opposite: to use facial characteristics and match them to DNA to determine if that face is a good representative of the DNA sample. They do this by breaking down the face, from a database of numerous faces from different backgrounds, into 63 different segments, after which they “score” each facial segment against the DNA based on characteristics such as sex, age, genetic background, and BMI, all of which can be inferred from a DNA sample. So for example, if the DNA is female (that is, has two X chromosomes), then a more typically feminine face shape would have a higher score than a more typically masculine face shape. Or if the DNA is from a certain ancestral background, then facial shapes corresponding to that background would have a higher score than other face shapes. They score each segment individually and independent to other segments, and they add all the scores together to achieve a single matching score. The higher the score, the higher the probability that a particular face matches a particular DNA sample.

Facial segmentations for the global sample (left) and the European sample (right). Both begin with the full face as segment number 1, and each previous segment is further partitioned into two other segments. This partitioning is done 5 times, for a total of 63 different facial segments.

But where do they get these facial images to match against the DNA from? For the study itself, scientists selected individuals from a global sample and a European sample with their informed consent. The global sample is used to identify a person’s genomic background within diverse populations, while the European sample is used to determine facial shape variation related to individual genetic variation. That is, the global sample can be used to identify the population in which the DNA belongs to, while the European sample can be used to better hone this technology and pinpoint an individual from a specific population. Should this method of determining facial characteristics from DNA be implemented in the future, then there are numerous places and sources where someone’s face can be used as a reference – driver’s licenses, passports, other governmental databases, etc. Understandably, since these images would not be public information but rather government-handled, this would lead to moral and ethical debates – would this be considered a breach in privacy? Is this even legal?

The issue of privacy is not the only thing of concern. As of now, it is not feasible to expect this method to regurgitate one single face that best matches the DNA. Instead, the goal of this technology is to pinpoint faces that would look a lot like the person you want to find. Since it is looking through a database, if that database contained only a few minorities, then this opens the possibility of potential bias. Since matching scores are based only on probability, if that result turns out to be wrong against the odds, then this could lead to incriminating an innocent person.

It would still need a lot of work including scientific critiques, implementation of legally binding regulations, and further adjustments to the database and algorithm in order to prevent biases. Despite these, there is definitely potential in facial recognitions through scoring of different facial characteristics with genetic bases, and I look forward to how this method can be applied in the future.

Learn more about face prediction from DNA

1. How to read the genome and build a human being (TED video)
2. How far scientists go to create your face from your DNA (YouTube video)

About the Author

Guia Janelle Pucyutan is set to graduate this year with a specialization in Human Biology. She has always loved to study genetics, and she hopes to work as a medical laboratory technologist in the future. When she is not bombarded by assignments and readings, she likes to read, play video games, listen to kpop, and cook!

Featured image: Sharon McCutcheon on Unsplash (license).