Do you believe it? Genes can "predict" a person's appearance

I saw an interesting question: If we take a person's DNA sequence as input and a frontal photo as output, and throw it into a deep neural network for learning, is it feasible?

Of course it is feasible, and some people are already doing it. For example, everyone has heard of "genetic identification of people", and today I want to talk to you about it.

01. Has the genetic identifier arrived?

J. Craig Venter, a great biology figure whom I introduced before, a man with both good and bad reputations, did this.

The godfather of gene sequencing published the results of a face prediction software. He sequenced the entire genome of people of different ages and ethnic backgrounds, and then input the genetic data into a computer.

The age composition of the research group can be said to cover most age groups.

The ethnic composition of the research subjects basically covers the mainstream ethnic composition. For example:

Ancestry components are African (AFR);

Native American (AMR);

Central South Asian (CSA);

East Asian (EAS);

European (EUR)

Gene sequencing data constitutes Inferred genomic ancestry proportions for each study participant.

Gene identification, this great innovation, is about to start...

02. How did he do it?

Next, the individual’s appearance information is input, such as three-dimensional facial structure, voice characteristics, biological age, height, weight, BMI index, eye color, skin color, baldness or hair color, etc. [1]

The next step is to train it using the newly developed algorithm, which is the machine learning that everyone is familiar with.

What was the result, did he succeed?

03. What is the result?

Based on their genetic data, as well as high-quality 3D photos of the participants, their algorithm can make very successful predictions of people's appearance. The following figure shows the pixel changes of genetic predictions and facial 3D structure

Model Training

Here are the results: The picture below is the result of genetic prediction. The left side is the real face of the person, and the right side is the result of genetic prediction.

For example, they randomly selected a group of people (10 people) from a public database Human Longevity (HLI) for identification, and the accuracy rate was as high as 74%, which exceeds the proportion of most criminal investigations.

You know, there was a study that asked eyewitnesses to describe the appearance of criminals, and the results were terrible. For example, I found a study [2]

Objective To verify the accuracy of the simulated image and computer portrait combination technology of the suspect through the description of the witness. Methods The hairstyle, chin, eyes and other facial features of 15 male and female photos were removed separately, and one of them was identified as the identification target (object) and identified by 786 witnesses. Results The overall accuracy rate of eyewitnesses in identifying facial features was 17.4%.

The accuracy rate is only 17.4%, which is basically meaningless. It can be seen that genetic prediction is awesome. It's not just about face, there are other factors, such as age.

For example: eye color

For example: height, weight, BMI

For example: skin color

He also gave the prediction accuracy of this model. Although the accuracy of different indicators varies, it is still quite high overall.

This is the real blood test.

You know, everyone must be familiar with it. It was very popular among the ancients to identify relatives by dropping blood. Of course, modern science has denied this, because blood identification has a high failure rate and can be falsified through technology, so poor ancient people who were not recognized.

04. What is its use in real life?

Everyone must be familiar with finding criminals through DNA traces in their blood. However, this is actually quite difficult because you must find a comparison object. Otherwise, even if you obtain the criminal's DNA, you can only wait foolishly for one day in the future when you are lucky enough to find the criminal's DNA again, and then compare the two online. If you can no longer collect the criminal's DNA, then you will never be able to find the criminal through genetic evidence.

However, using DNA to directly predict a person's appearance is much more sophisticated, and it also leads to a core issue: privacy.

So when this article was published, it caused a lot of controversy. Some people thought the article was wrong, and some people thought it violated privacy. But in any case, it was a huge discovery.

In the future, as more and more samples are collected and more and more training is provided, it will be very likely that people can be predicted directly through their DNA.

【1】Lippert, Christoph, et al. "Identification of individuals by trait prediction using whole-genome sequencing data." Proceedings of the National Academy of Sciences114.38 (2017): 10166-10171.

【2】Research on the ability of eyewitnesses to identify facial features[J]. Criminal Technology, 2001(05):20-27.