Machine Learning of DNA Predicts Height Within an Inch

An algorithm can predict someone’s height based off their DNA, a new Michigan State study presents.

The new tool predicted thousands of peoples’ heights to within about an inch—and also showed predictive capabilities of bone density and even some educational attainment.

With a large-enough population, plus the time for machine learning to pick apart incredibly complex markers in entire genomes, the algorithm could determine further traits, according to the paper in this month’s Genetics.

The conclusions appear to offer forensic promise, as previous phenotyping advances like hair and eye color have contributed to some criminal investigations in recent years.

“Traits vary by heritability (the degree to which they are controlled by genetics), but to first approximation almost everything from eye color to height to facial morphology will eventually be predictable from DNA once enough data is available for analysis,” Stephen Hsu, vice president for research and graduate studies at Michigan State, told Forensic Magazine.

The dataset was massive: some 500,000 adults whose DNA is cataloged in the United Kingdom BioBank.

Some 453,000 of those genomes were used for “training runs”—so that the machine could look at hundreds of thousands of SNPs, and compare countless variations and linkages to the height of each participant, the study states.

Some 20,000 key SNPs were identified as impacting height among the population, according to the computer analysis.

The correlation was tested on five separate non-overlapping sets of 5,000 individuals apiece.

Their results: everyone’s height was predicted to within about an inch, based solely off the machine-learning interpretation of their genetic markers.

“The algorithm looks at the genetic makeup and height of each person,” said Hsu. “The computer learns from each person and ultimately produces a predictor that can determine how tall they are from their genome alone.”

Bone density at the heel and the educational attainment were also checked against the huge database in a similar methodology. But it was not as precise as height. But it was effective for determining outliers, like those with very low bone density like that shown in osteoporosis, and also with those with significant struggles in their school years.

Further outcomes and traits, from disease to appearance, could be cataloged and predicted, based off artificial intelligence scrutinizing the minutiae of whole genomes, according to the paper.

“We are optimistic that, given enough data and high-quality phenotypes, results similar to those for height might be obtained for other quantitative traits, such as cognitive ability or specific disease risk,” the authors write.

“This is only the beginning,” added Hsu.

Phenotyping has shown limitations so far, however. Susan Walsh, a scientist at Indiana University–Purdue University Indianapolis who unveiled a new webtool for appearance prediction earlier this year, told Forensic Magazine at the time that predicting something as detailed as a facial appearance is not yet logistically possible, with the current state of DNA science.

“We are nowhere near understanding facial prediction,” said Walsh. “At the end of the day, there’s no genetics behind it … Pigmentation—absolutely.”

By Forensic Magazine

Leave a Reply

Your email address will not be published. Required fields are marked *