As word spread in 2018 about the birth of the world's first genetically altered babies, concerns over their future health mounted, with one study even raising the tragic possibility of shortened lives for the newborns. That risk now seems far less likely.
The alarming paper published in Nature last June has now been retracted by the authors themselves, who in the wake of criticism admit the way they searched for signs of a mutated gene in a data sample left too much room for doubt.
It's an important lesson not only in how science values self-correction, but how researchers need to tread lightly as they trawl through population-sized databanks in search of new discoveries.
"I feel I have a responsibility to put the record straight for the public," University of California population geneticist, Rasmus Nielsen, told Ewen Callaway at Nature.
The gene at the centre of the research serves as a template for a receptor on white blood cells.
Called CCR5, its usual job is to detect chemical signals used in immune responses. Unfortunately the deadly human immunodeficiency virus ( HIV) evolved to use it as a window to gain entry into the cells.
Ever since the receptor's role in HIV infection was discovered, researchers have wondered just how important this receptor really is. Would we really miss it if it was gone?
Luckily an answer might be found among a percentage of people of European descent with a naturally occurring 'broken' version of CCR5 called delta-32. Those who carry a single copy of the delta-32 variant seem to be less susceptible to HIV than the rest of the population.
He Jiankui's initial announcement suggested at least one of the twins was carrying two altered CCR5 genes. While they don't appear to match the delta-32 variants, it was enough to invite speculation over what kind of lives the children might have.
HIV resistance is no doubt a good thing in a world where the disease it causes is still destroying too many lives. But those benefits to any one individual might not be so great if a low quality CCR5 receptor raises the risks of developing other health problems.
Nielsen and his colleagues intended to answer this question by looking for similarly altered versions of the CCR5 gene in the UK Biobank's giant genetics database.
They estimated about 1 percent of the records in the database came from individuals with two delta-32 variant copies of the gene. Importantly, they calculated that this tiny fraction was 21 percent more likely to die before their 76th birthday, compared with those at least one 'normal' copy of the gene.
Thankfully, as happens in science, big claims often attract sceptical inquiries. Others quickly dived into the statistics in search of similar correlations using both the UK Biobank and other nation's datasets, coming up empty handed.
So where did Nielsen and his colleagues go wrong?
The cause of the discrepancy could lie in how the data was collected in the first place.
One way to work out whether a person has a specific gene is to simply use a template that sticks to a target sequence. These probes don't always work perfectly, meaning some people will incorrectly appear as negative in the database.
By potentially undercounting the number of people with the CCR5 delta-32 receptor, Nielsen risked masking the true impact of the mutation, making it look like there is a difference in mortality statistics. Which is why he asked for the paper to be retracted.
For researchers, huge banks of genetic and medical data collected from across a population provide the necessary quantities of information needed to spot subtle patterns that demarcate healthy from unhealthy bodies.
Yet as potentially useful as those statistics are, there's dangers in forgetting they come with plenty of assumptions.
You can see the now-retracted paper here.