Four years ago, Candra Alston and her 3-year-old daughter, Malaysia Boykin, were murdered inside their South Carolina apartment. Police in Columbia collected DNA at the scene, but the investigation stalled.
Prime-time crime shows would have you believe DNA samples can convict the guilty and clear the innocent, but real life is more complicated. In order to find an assailant, the DNA has to match either a previous offender in the FBI’s CODIS database or a sample from one of the victim’s acquaintances. When it doesn’t, the investigation hits a wall. In the South Carolina case, police gathered 150 DNA samples and conducted 200 interviews with likely suspects — and still nothing.
Now the Columbia police are experimenting with a new technology that uses tiny amounts of DNA to create a computer-generated illustration of their suspect. Snapshot, a program developed by a Reston, Va., company called Parabon NanoLabs, goes beyond simply listing physical attributes — eye color, hair color, ethnicity and facial features — and creates a 3-D image of what the killer might look like. The police in South Carolina hope that publicly releasing the suspect’s image and description will bring up fresh leads in a stale case.
Dabrien “Dabe” Murphy, the senior solutions architect at Parabon, sits in front of three monitors in a little office on the fourth floor of an unremarkable Reston office building — not exactly the first place that comes to mind when you think of a lab. With a few keystrokes, Murphy brings up a revolving 3-D image — the back of a head. Another few taps and a face attaches itself along the hairline. The face is a man’s: olive skin, greenish eyes, full lips.
Murphy has fed DNA markers, linked to certain facial attributes, into 3-D imaging software to create what he calls a “composition.” It produces a somewhat distorted image where the face meets the rest of the 3-D model.
“There’s a little bit of, okay, manual manipulation to make this not look quite so Frankensteined,” Murphy explains. Using his cursor to adjust points and axes in the imaging software, Murphy smooths out the hairline and jaw where the projected face attaches itself to the head.
According to the markers Murphy feeds into the imaging software, the man on the screen is of Northwest African ancestry, with hazel or green eyes; black or brown hair; and few or no freckles. This man’s DNA was publicly available, so they’re using it to test their model; they know, from the data included with the DNA, that he is Algerian. The trait predictions come with varying levels of confidence; Parabon’s scientists are 73.4 percent sure that their skin color prediction is accurate, but they are 94.7 percent confident in the subject’s eye color.
All this from 9.6 nanograms of DNA. That’s less than 0.00000001 grams, an amount so small, it’s hard to compare to anything else.
Snapshot combs through a genotype, searching for significant markers and clusters that might indicate physical attributes and removing unimportant variables.
“We start from our million markers and then we cut that down to hundreds or thousands,” says Ellen McRae Greytak, the bioinformatics manager at Parabon. “We’re not specifying ahead of time markers that we think might be important. We’re just searching through the whole genome to find what is important.” Greytak and others then cross-validate their prediction against DNA samples not included in their model.
Greytak and Steven Armentrout, a Parabon founder and chief executive, say they have performed sample analyses for major metro police departments and federal agencies, and they are working on about a dozen cases. But Parabon’s scientists won’t reveal which agencies they’ve worked with, and they would not show these images and photographs of sample subjects to The Post, citing privacy issues. The only publicly available image is the South Carolina suspect — a case that is still unsolved.
Moses Schanfield, professor of forensic science at George Washington University, is skeptical. “How many times has this led to an arrest and conviction?” he asks. He is wary of latching onto the latest shiny innovation. “We’ve got lots of things we can do technologically. Whether or not they in fact work is a whole different question.”
Schanfield also questions Parabon’s decision not to publish any of its work. “So none of it’s peer-reviewed?” he asks. “There’s no track record. There’s no data. There is nothing hard that establishes that they can actually do what they say they’re doing.”
It’s your high school math teacher’s favorite adage: Show your work.
“We’re not trying to publish a paper,” Armentrout counters. “We’re trying to do prediction.
“This is for generating a lead. We’re providing information to help the law enforcement community be more effective.”
Any lab can test for eye color, hair color and ethnicity to some extent, but Greytak says Snapshot offers greater detail and higher levels of confidence in the accuracy of its predictions. Greytak is also confident that the program can pinpoint ancestry in greater detail than before. “So, someone has one parent from one place and one from another, we’ll detect that,” she says. “That’s very new.”
“There’s a lot of information in DNA that is just not being used in the forensics field.”
Predicting face shape and features is also new. Snapshot examines the effects of characteristics such as sex and ethnicity. “The most important one is going from a very feminine face to a very masculine face,” Greytak says. “This affects the whole face: the cheekbones, the length of the face, the shape of the nose.” They don’t expect to find markers for a single feature, such as where the tip of the nose falls. “But there might be a marker that does affect the whole size of the nose.”
The company is developing Snapshot through a small-business grant from the Department of Defense. It works with outside labs to do the genotyping.
Snapshot is being used on cold cases such as that of Alston and her daughter, but it could be used in any case with DNA samples.
“By narrowing the suspect pool and allowing investigators to focus on high-value suspects, we aim to keep cases from going cold in the first place,” Greytak says.
Armentrout says he thinks this technology could one day be applied to less-major cases, such as property crimes. “We think it’s game-changing,” he says.
Predicting, with a certain level of confidence, a subject’s features could be helpful in an investigation. However, Parabon’s scientists have learned that excluding certain characteristics may be even more important.
“When we got into this, we only gave point predictions: ‘Skin color is X,’ ” Armentrout says. “We quickly learned that what they [law enforcement] really prefer is a higher-confidence statement about what it’s not.”
Of course, there are other characteristics not determined by genetics, like age and weight. And while researchers have found markers for height, no one has yet determined how one’s DNA actually affects how tall they’ll be.
Creating the images, too, has come with certain challenges. For instance, creating realistic 3-D hair — including facial hair — has been quite difficult.
“We’re still working on it,” Armentrout says.
Schreiber is a writer who works as a program manager of the International Reporting Project in Washington.