Composite Faces from DNA Help Solve Cold Cases

But now, nearly a quarter-century later, Montgomery County police have turned to a tool the department has used in previous cases, in hopes of generating new leads that may lead them to the scientist's killer. They hired Parabon NanoLabs, a Virginia-based technology company that has developed a technique called DNA phenotyping. Utilizing the genetic evidence, Parabon predicts the ancestry and physical appearance of an unidentified suspect, such as the likelihood that a person has a certain skin, eye or hair color, while excluding other possibilities. An artist then creates an image suggesting how the suspect might possibly look. In this case, for example, Parabon provided a possible likeness of the suspect in 1994, assuming he was age 25, and a second, age-progressed image of how the same man might look decades later.

"When people see these composites, it enables them to think back on anything that may have been unusual around the time of the murder and if we should take a look at anything specific," Sgt. Christopher Homrock, supervisor of Montgomery County's Cold Case Section, says in an email. "It also lets us reengage current and retired officers who were working in the area of the crime in an attempt to jog their memories."

"We always keep in mind that the composite is an estimation of what the suspect may have looked like. We try not to focus on finding exact matches when searching for possible suspects," Homrock says. "But the composite definitely allows us to go in the right direction."

In recent years, scores of police departments across the U.S. have used Parabon's DNA phenotyping in an effort to crack difficult cases, according to the company's website. In Newport Beach, California, for example, police hope it will generate new leads in the 1973 sexual assault and murder of 11-year-old Linda O'Keefe, who was abducted as she walked home from summer school. Parabon's predictions about the suspect's appearance and ancestry are included at the end of #LindasStory, a string of Twitter messages that the department posted in an effort to jump-start the investigation.

Newport Beach Police Sgt. Ryan Peters says that after the images of the suspect's possible appearance were released, the department received more than 200 leads of various types, though he cautioned that not all of them might prove useful. But with a 45-year-old unsolved case, "we need to try something to keep it alive," Peters says.

Ellen McRae Greytak, Parabon's director of bioinformatics, says that Parabon first began working upon predicting appearance from DNA while vying successfully for a contract for the U.S. Department of Defense a few years ago. While other companies' proposals focused on a few traits, such as eye color, Parabon argued that with enough data, other physical characteristics could be predicted as well, she says.

To develop a basis for predictions, the company turned to approximately 15,000 to 20,000 anonymous volunteers, who provided their DNA and also filled out detailed descriptions of themselves. Some of those subjects — less than 5,000, according to Greytak — also had 3D images taken to show their facial shape as well.

The company data-mined that lode of information, and then used machine learning to figure out how to predict a person's appearance. "We take the parts of the DNA that we find are important for each trait, and put them together into a predictive model," Greytak says. "And that [model] can be applied to new data. If you put in the DNA of a new person, it will make a prediction of the traits."

When Parabon began offering the service to law enforcement agencies in 2014, it initially encountered skepticism, since the idea of predicting a person's appearance from DNA "sounds so science fiction," Greytak says. As a test, police departments sometimes sent samples of DNA from a volunteer in their offices, and asked the company to predict that person's appearance. The results were good enough to convince them to give it a try, Greytak says.

Since then, police departments across the U.S. have used Parabon's services as an investigative tool.

It's important to note that the profiles only give the likelihood that a suspect has certain characteristics, and police can't actually use them to make a conclusive ID. "Phenotyping is not going to point to a single individual and it's not going to point to an arrest or a conviction," Greytak emphasizes.

But investigators can use the probability of a suspect having certain facial characteristics and ancestry to go through a large list of suspects and identify a few to prioritize, Greytak says. She notes that there's almost always a trait that can be excluded with high confidence. "We tell them to pay the most attention to high confidence-level traits and low-likelihood ones," she explains.

"Getting information based on a composite is only the first part of investigating that lead," Homrock said. "If someone remembers a name associated with that composite, we can do things like look at criminal histories, complete timelines on the individual, interview acquaintances, or interview the person. The composite only points us in a direction; it's up to the investigator to determine with more work if the person is responsible for the crime or can be eliminated as a possible suspect."

"We don't jump to conclusions based only off of a tip generated by a composite," Homrock continues. "The tip from the composite is only a small part of the investigation. The work by the investigators that comes after the tip is a much larger part. This work either confirms that the person is responsible or eliminates them as a suspect. The composite is definitely an important part, but only one of many parts of the entire investigation."

Nevertheless, in one rape-homicide case in Costa Mesa, Calif., a police lieutenant told the Los Angeles Times that there was an uncanny resemblance between the Parabon image and a suspect whom police identified separately through fingerprint evidence from the scene.

DNA phenotyping also has its critics. American Civil Liberties Union policy analyst Jay Stanley, who wrote this 2016 blog post about DNA phenotyping, says that he's concerned that using the images based upon predictions might alter witnesses' memories, leading them to errantly identify a person who looked like the picture.

But Newport Beach's Sgt. Peters says that such a tip wouldn't be sufficient in itself to make an arrest, and that much more proof would be needed. "Someone might say to us, there's a guy in the neighborhood who resembles that picture, but I didn't think anything of it at the time," he explained. "Frank from the neighborhood is worth looking at. But our job is still to go out and get his DNA and compare it to the DNA from the crime scene."

"This just gives us a little better sense of direction," Peters says.