One of my guilty pleasures is true crime television shows. I’m fascinated by the twists and turns that go into identifying, accusing, and convicting a criminal and the mystery surrounding those who have (so far) gotten away with their crimes.

Several years ago, a new forensic technique called DNA phenotyping caught my eye. DNA phenotyping estimates the appearance of suspects whose DNA was found at the scene of a crime. It’s an intriguing idea, but I think this technology has limited use for law enforcement.

A local example illustrates the limits of this technology. In 2010, Sherry Black, the mother-in-law of a prominent Salt Lake City businessman was murdered in the book store she ran. Her killer was unknown, but he did leave behind a DNA sample. In 2017, Parabon NanoLabs released an estimate of what the man might look like, based on DNA found at the scene. The results are below.

DNA phenotype results from Parabon NanoLabs estimating the appearance of the killer of Sherry Black in Salt Lake City, Utah, in 2010. Source.

In addition to the composites above, Parabon NanoLabs estimated that the killer would have 51.33% African ancestry and 34.19% northern European ancestry, black or dark brown eyes, light brown skin color, black hair, and few or no freckles.

In October 12, 2020, a suspect was arrested: Adam Antonio Spencer Durborow. If he is the killer, then this give us the perfect opportunity to compare the DNA phenotype results to the accused man’s actual appearance.

Left: Estimated appearance of Sherry Black’s killer, based on DNA phenotyping from Parabon NanoLabs. Right: Booking photo for accused murderer Adam Antonio Spencer Durborow, who was 29-years-old at the time the photo was taken. Mr. Durborow is presumed innocent until proven guilty.

It is not a match. Beyond a few vague similarities (nose width, hair color), these don’t look like the same person at all. (The skin color looks very different, but this might be an artifact of the booking photo’s lighting.)

The problem with the face generated from the DNA is that it is so generic. Parabon NanoLabs has made similar comparisons available, and they are not impressive. Within each race and sex group, the faces look . . . average. Some of them look very little like the real person (e.g., the Yolanda McClary example), and when the match is close, it is because the real person has a generic-looking face for their race and sex (e.g., the Andrea Canning example).

This should not be surprising. Most polygenic scores derived from DNA have a very low correlation with phenotypes. As a result, there is going to be a large amount of regression towards the mean for almost every trait. The result when making predictions from DNA will be a face with nearly average chin shape, average cheekbone height, average face width, average hairline, etc.

This is why I do not think that these estimated appearances are useful for law enforcement. Indeed, many of the faces look like too many people–and often not the person that the DNA belongs to. Using DNA to identify broad characteristics of a suspect–their sex, race/ethnicity, and perhaps approximate height–is probably useful. Beyond that, I wouldn’t put any trust in these images.

The real 21st century breakthrough in using DNA to identify criminals has been forensic genealogy. This technique uses public DNA databases to match DNA samples to distant relatives. Genealogical records are then used to identify potential people that the unknown DNA could have come from. (A new DNA sample from the known suspect confirms or disproves this identification.) This has already been used to identify Joseph James DeAngelo (the Golden State Killer), serial killer Terry Rasmussen, and many other criminals.

As far as I know, DNA phenotyping has never successfully identified a criminal. I do not think that will ever change. The technique seems to be a forensic dead end.

css.php