But this white one is based solely on my DNA.

That's not what I expected at all.

It's not an exact match, to say the least.

But forensic scientists are tirelessly

improving their ability to identify people

without ever seeing their faces.

Why?

In criminal cases involving unidentified remains,

recognition can be the last chance to develop leads

after traditional methods fall short.

And if they're lucky,

giving a victim their identity back could close the case.

That's what happened in 2017

when a company called Parabon

created this facial reconstruction.

The image got 26-year-old Shaquana Caldwell recognized,

and, in turn, caught her killer after months with no leads.

Parabon combined two methods of facial reconstruction,

forensic sculpture and DNA phenotyping.

Forensic sculpture uses skull remains

and population samples to create an idea

of what the victim looked like.

But this method won't ever be 100% accurate.

DNA, on the other hand,

could get us as close as we can to a copy machine.

That's how we ended up at Shriver Lab at Penn State.

Sarita: This is just envelopes and envelopes of hair.

Abby: In any other place,

this cabinet full of hair would be really weird.

They're collecting data about

people's hair type, facial features,

and more to build an algorithm

to take my DNA

and spit out a replica of my face.

Producer: Does everyone sound like this when they do it?

[all laughing]

Sarita: No! I've never heard anyone sound like this.

I'm nervous I'm not going to recognize myself

when I see it on the screen,

not because it's not a good reproduction,

but because I

have a different vision of myself than reality.

Mark: Ready? Abby: I'm ready.

OK! That's not what I expected at all.

This final image is a composite

of my genetic info and the collected data

from people who share my genetic ancestry.

Right now, the algorithm works best,

meaning the predicted face is most recognizable,

when a person looks, well,

statistically average.

So, you can get ancestry axes

estimated from the genotypes,

and then we layer on top

the feature-specific genotypes.

Abby: Mm, gotcha.

There are some big differences

between my actual and predicted faces.

Mark: Which is telling me that

we need more information in the model.

Abby: Humans share about 99.9% of DNA,

but what makes me look like me

is in the last 0.1%.

And that's thanks to single-nucleotide polymorphisms,

the variations in base pairs throughout our genes.

This is easiest to see with our eyes.

I have two adenine at the base pair,

most associated with having brown or hazel eyes.

Which I do.

Now, if you take the A's out,

and you replace them with two guanine,

then that likelihood shifts

towards blue or green.

Mark and his team are building a database

of which SNPs affect which features.

Mark: We're really at an early stage.

I mean, we have 200 genes

that have significant effects on the face.

Abby: At the moment, their formula isn't perfect.

A lot of traits are controlled by multiple SNPs,

so it isn't a simple one-to-one connection.

You know, the distinctive features

that we really rely on to identify each other

are not necessarily coming through.

Abby: Small details are missing,

like my cheek mole.

And my nose is a bit pointier in reality.

I don't usually think I look like my dad,

but without the mapping,

the texture mapping, I think,

this looks a lot like him.

Then those are probably genes with big effects,

because, you know, those differences that you see

that make somebody distinctive

often stand out in their siblings.

Like you said, your face looks like your dad.

Abby: Plus, SNPs are associated with different traits,

not definitive proof of them.

Like, I have all the right SNPs

for not having a cleft chin,

but tell that to all of my bullies

who called me butt-chin growing up.

With so much variation,

to get this closer to this,

Mark and his team need data from as many people as possible.

And so if you already have our DNA,

what are we getting here?

Sarita: In order to inform our research,

we need to take measurements and data on many phenotypes.

The idea is that, hopefully,

by collecting all of this data

and associating it with what you really look like,

we can then use that to inform

an algorithm to predict on someone

that we don't know this information about.

Abby: So I decided to offer up my data to the algorithm.

All right, snip away.

Hair can make or break a case,

but the most common form of forensic hair analysis

isn't scientific.

Sarita: The most common forensic form of hair analysis

often uses racial categories to categorize hair.

And so, in this way,

we're able to study hair more objectively.

So we're going to embed your hair

in a low-melt-point plastic

and then cut that embedded chip in half,

and that'll allow us to look down the shaft of the hair.

And so we call that cross-sectional.

And we can look at the elliptical nature of your hair,

whether it's more circular, the thickness,

see if you have a medulla,

because some people don't have a medulla in their hair.

Abby: What is a medulla?

Sarita: The medulla is the center portion of the hair.

You can kind of think of it as like

the bone marrow of the hair.

This is really important forensically

because if you are a forensic scientist,

you get up on the stand, you say it's an African hair type,

then you're going to sway the jury into thinking

that the perpetrator was a specific race.

Abby: They also sample your skin pigmentation.

Javier: So, what this does is it measures

the melanin in your skin.

I'm going to need three parts.

I'm going to need this part of your arm right here,

with your palm facing up,

because it's not exposed to a lot of light.

Abby: Oh, yeah, I hide that. Javier: We need your forehead,

because that's exposed to a lot of light actually.

And we're going to need your hair,

because we're going to be also looking at

your hair pigmentation.

Abby: This little thing is sort of like

what they use in makeup stores.

Javier: Yep, that's right.

Abby: For weight and BMI,

the Tanita scale shoots a small amount of electricity

through your body.

Sarah: OK, we're going to electrocute you.

Abby: OK.

Sarah: Small. Abby: What about my mic pack?

Sarita: It should be fine.

Abby: OK.

Oh, no. Sarah: You don't feel it.

Abby: I know, I'm just nervous.

Because I'm a participant in the research,

they need to know what my actual face looks like.

Sam: So, this is essentially our low-tech

zero-gravity machine.

[both laughing]

Oh, I feel it. I'm in space.

Gravity affects how your face lies on top of your skull.

So getting photos from all angles

provides a more accurate final product.

Sam: Basically what we'll do is we'll take all the photos

from all the angles that we took,

put them in to our vector software,

and actually stitch them together.

So we'll get something that you can actually

revolve around on a computer screen.

Abby: You're going to put this up next to

the recreation from my DNA

and see how they match or not?

Sam: Yeah, we're going to see how accurate

we can predict your face.

Abby: By using known faces to check their work,

they can hone the accuracy of the algorithm

when predicting unknown faces.

And the more times they see SNPs correlating

with different traits,

the more definitively they can connect them.

We don't need to sample everybody

to understand how something works.

We need a representative sample ethnically,

or population appropriate.

I wouldn't take our methods and the data we've collected

and go to Finland and say,

OK, let's set up a big phenotyping lab here,

without sampling Finnish people.

Abby: There's still a long way to go

before this technology can become mainstream.

One of my deep concerns over the forensic applications

of molecular photofitting

are the premature deployment of the methodology.

It's important that investigators

understand what you found

and will use that information that you found appropriately.

Otherwise, you give the investigator

an image of what the person might look like

and they can say, "Well,

that image is not the same as my defendant."

Abby: The team needs to collect more data,

improve the computational power of the algorithm,

and simply more time to get things right.

But even then, DNA is just your blueprint,

not the final product.

As we mentioned, SNPs rely on probability.

Plus, there's plenty of environmental factors

that come into play.

We don't expect every feature is predictable.

Even identical twins.

Abby: It can't tell us things like a person's gender,

if they have scars, or if they dye their hair

or wear makeup.

Some might enjoy the sun much more than the other twin,

and you can see those kinds of differences emerging

because of the environment.

Abby: The key to using DNA phenotyping to solve crimes

isn't perfecting the tech.

It's understanding its limitations.

This has been a 20-year project already,

and, you know, I expect will continue

in the next five, 10 years.

We'll have more than double our current sample size

and be able to say a lot more about appearance.

All right. Abby: Thank you so much.

Sure.

Do I have a Medusa?

Medulla, sorry.

Sarita: Oh, is that what you were trying to say?

Abby: That's what I was trying to say.