Fifty of the world's most widely used genetic prediction programs consistently overestimate the danger of rare mutations while underestimating the risk of more common ones, according to a new study that tested the tools against 13.5 million mutations across 6,659 human genes.
For the study, published in the American Journal of Human Genetics, researchers at the Centre for Genomic Regulation in Barcelona, found the bias stems from how nearly all the software programs handle DNA.
Programs like AlphaMissense, EVE and popEVE check whether a DNA region has remained unchanged across millions of years of evolution. If it has, the software assumes the spot must matter and that altering it is likely harmful. But the researchers found that some regions of the genome are naturally far more mutation-prone than others. For example, previous research by the group has previously shown that the starting points of genes are 35% more prone to mutations than other regions—and the prediction tools don't account for underlying variation in mutation rates.
To test the tools, the team used experimental data from the lab of Ben Lehner, also at the CRG, whose group had physically introduced hundreds of thousands of mutations across 500 human protein fragments and measured the resulting damage.
The results showed mutations that occur more frequently tend to be slightly less damaging—evidence that life has built up some tolerance for its own most common errors, an idea long proposed in theory but never previously confirmed through direct measurements of protein function. Even after correcting for that effect, the researchers found it was too small to explain the bias seen across the 50 prediction programs.
The consequences cut both ways: the study found mutations in genes for DNA repair, cilia and sperm function are likely being flagged as more dangerous than they are, while mutations in genes linked to intellectual disability and conditions inherited from a single parent are likely being underestimated.
Data from Center for Genomic Regulation