Criticism for Craig Venter’s “Face-Prediction” Software

The tool cannot predict a person’s face from DNA, say researchers—including some listed as coauthors on the publication. 

Sep 11, 2017
Catherine Offord


Update (September 12): Venter and his colleagues have published a response to Yaniv Erlich’s criticism of the paper. On bioRxiv, the authors refute Erlich’s claim that the paper contains “major flaws,” and call his approach to the data “misleading.” They also emphasize that their software can predict not only faces from DNA, but other phenotypic and demographic variables such as height and biological age.

A paper from Craig Venter and colleagues at his genetics company Human Longevity made waves after publication in PNAS last week (September 5) with its claim to be able to reconstruct a person’s face from his or her genome. Now, it’s attracting a tidal wave of criticism from researchers—including some who are listed as coauthors—with suggestions that the article is more self-serving for Venter’s private company than it is a scientific advance.

“Craig Venter cannot predict faces,” writes Yaniv Erlich, a computer scientist at Columbia University, in a post today (September 11) on figshare. To illustrate his point, Erlich, who has also penned a longer critique on bioRxiv, created a panel of faces: one of Craig Venter, another that the new software predicted using Venter’s genome, and a third of actor Bradley Cooper. “In short, many white males in the same angle would look like the predicted face,” he writes.

The company claims the software uses DNA to predict facial features, from skin color to cheekbone height. While the concept itself is not that far-fetched, the software does not come close achieving its goal, says Mark Shriver, an anthropologist at Pennsylvania State University who reviewed the paper for Science, where it was ultimately rejected. “Calling it predicting from the genome is what’s wrong,” he tells MIT Technology Review. “The main message is way overstated.”

One theory from critics is that the software relies mostly on race and sex to make predictions about average face types—a result backed up by the software’s poor performance on relatively homogeneous sets of data. From a set of 20 photographs of men and women of different races, the software picked out the correct face using DNA around 70 percent of the time. That rate dropped to 11 percent when only male European faces were included.

The paper argues on the basis of its results that access to DNA databases should be restricted for privacy reasons, raising a conflict of interest for a company that is building a giant genetic database, says study coauthor Jason Piper, who has since left Human Longevity to work at Apple. “I think genetic privacy is very important, but the approach being taken is the wrong one,” he tells Nature. “In order to get more information out of the genome, people have to share.”

A spokesperson for Human Longevity tells Nature that although the sample size of 1,061 people overall was small, the company stands by its methodology. What’s more, “the [Human Longevity] team is working on rebuttal to criticisms by Yaniv in BioRxiv.” 

September 2018

The Muscle Issue

The dynamic tissue reveals its secrets


Sponsored Product Updates

StemExpress LeukopakâNow Available in Frozen Format

StemExpress LeukopakâNow Available in Frozen Format

StemExpress, a Folsom, California based leading supplier of human biospecimens, announces the release of frozen Peripheral Blood Leukopaks. Leukopaks provide an enriched source of peripheral blood mononuclear cells (PBMCs) with low granulocyte and red blood cells that can be used in a variety of downstream cell-based applications.

New Antifade Mounting Media from Vector Laboratories Enhances Immunofluorescence Applications

New Antifade Mounting Media from Vector Laboratories Enhances Immunofluorescence Applications

Vector Laboratories, a leader in the development and manufacture of labeling and detection reagents for biomedical research, introduces VECTASHIELD® Vibrance™ – antifade mounting media that delivers significant improvements to the immunofluorescence workflow.

Enabling Genomics-Guided Precision Medicine

Enabling Genomics-Guided Precision Medicine

Download this eBook from Qiagen to learn more about the promise of precision medicine and how QCITM Interpret can help deliver better care with better knowledge.

Best Practices for Sample Preparation and Lipid Extraction from Various Samples

Best Practices for Sample Preparation and Lipid Extraction from Various Samples

Download this white paper from Bertin Technologies to learn how to extract and analyze lipid samples from various models!