A recent study led by Dr. Donate Weghorn at the Center for Genomic Regulation (CRG) in Barcelona suggests that some tools used to predict the harm caused by genetic mutations might not be as accurate as previously thought. These tools are used to assess whether a genetic change could lead to disease or other health issues. The research team tested 50 of the most widely used prediction programs by analyzing 13.5 million mutations across 6,659 human genes. Their findings, published in the American Journal of Human Genetics, revealed that most of these tools overestimate the risk of rare mutations and underestimate the risk of more common ones. The researchers found that the tools often rely on the assumption that regions of DNA that have remained unchanged over millions of years are important for survival and that any change there is likely harmful. However, this approach can lead to bias. For example, tools like AlphaMissense (developed by Google DeepMind) and EVE and popEVE (developed at CRG) were found to have this same tendency. They give more weight to mutations in regions that are less likely to change, while underestimating the impact of mutations in regions that are more prone to change. The study also identified that some parts of the human genome are more likely to experience mutations than others. For example, the start of genes is 35% more mutation-prone than other regions. The researchers found that considering mutation rates could change how some genetic variants are classified, from harmless to harmful. However, for most cases, the change would be relatively small. Weghorn noted that this could have implications for understanding the actual risk of certain mutations. To test the accuracy of the prediction tools, the researchers used data from experiments conducted by Ben Lehner, also at CRG. Lehner's team created and tested hundreds of thousands of mutations in human proteins and measured their effects. They found that more common mutations tend to be slightly less harmful, possibly because evolution has allowed for some tolerance of these frequent errors. However, the effect was too small to explain the significant bias seen in the predictions made by the tools. The study recommends that future genetic prediction tools should take into account how mutation-prone different regions of the genome are. The researchers emphasized that their findings do not indicate how this affects individual patients, as these tools are only one part of the diagnostic process. Clinical experts use a range of information to make decisions about patient care.