Tweet
DeepMind, the advanced artificial intelligence division of Google, has introduced a tool named AlphaMissense designed to discern the impact of missense mutations on disease susceptibility. Missense mutations are alterations in the genetic code that lead to a change in the amino acid sequence of proteins. These mutations contribute significantly to the pool of “variants of uncertain significance,” where the consequences of the mutation are unknown.
While advancements in DNA sequencing have empowered scientists to pinpoint potential disease-causing genetic alterations, the challenge has been in accurately interpreting their effects. This interpretation issue is especially pronounced for missense variants. Although humans carry numerous missense variants, only a fraction have been authenticated as either harmful or harmless by experts.
In a bid to refine the classification process of these genetic changes, Jun Cheng and his team derived AlphaMissense from DeepMind’s renowned AlphaFold system, which predicts protein structures based on gene sequences. Rather than focusing on the structural shifts caused by mutations, AlphaMissense amalgamates databases of related protein sequences and the structural background of these variants. This fusion results in a score reflecting the potential of a variant to cause disease. Through this method, the team made predictions for an impressive 71 million missense variants.
Tests suggest that the DeepMind tool outperforms existing computational tools when it comes to distinguishing between benign and harmful variants. Moreover, when pitted against lab experiments assessing thousands of mutations simultaneously, AlphaMissense demonstrated robust capabilities.
To ensure a comprehensive analysis, the research team constructed a database detailing every conceivable missense mutation in the human genome. Their findings suggest that 57% of these are probably harmless, while 32% might be responsible for inducing disease.
Despite the promise shown by AlphaMissense, experts such as Arne Eloffson, a computational biologist at the University of Stockholm, believe its advancements are significant but not earth-shattering. Other experts caution that computational predictions should remain a supplementary resource when diagnosing genetic diseases.
Yana Bromberg, a bioinformatician at Emory University, emphasizes the importance of rigorous evaluation of such tools. Proper evaluation is critical before any real-world application to ensure that it offers trustworthy insights for health practitioners.
While advancements in DNA sequencing have empowered scientists to pinpoint potential disease-causing genetic alterations, the challenge has been in accurately interpreting their effects. This interpretation issue is especially pronounced for missense variants. Although humans carry numerous missense variants, only a fraction have been authenticated as either harmful or harmless by experts.
In a bid to refine the classification process of these genetic changes, Jun Cheng and his team derived AlphaMissense from DeepMind’s renowned AlphaFold system, which predicts protein structures based on gene sequences. Rather than focusing on the structural shifts caused by mutations, AlphaMissense amalgamates databases of related protein sequences and the structural background of these variants. This fusion results in a score reflecting the potential of a variant to cause disease. Through this method, the team made predictions for an impressive 71 million missense variants.
Tests suggest that the DeepMind tool outperforms existing computational tools when it comes to distinguishing between benign and harmful variants. Moreover, when pitted against lab experiments assessing thousands of mutations simultaneously, AlphaMissense demonstrated robust capabilities.
To ensure a comprehensive analysis, the research team constructed a database detailing every conceivable missense mutation in the human genome. Their findings suggest that 57% of these are probably harmless, while 32% might be responsible for inducing disease.
Despite the promise shown by AlphaMissense, experts such as Arne Eloffson, a computational biologist at the University of Stockholm, believe its advancements are significant but not earth-shattering. Other experts caution that computational predictions should remain a supplementary resource when diagnosing genetic diseases.
Yana Bromberg, a bioinformatician at Emory University, emphasizes the importance of rigorous evaluation of such tools. Proper evaluation is critical before any real-world application to ensure that it offers trustworthy insights for health practitioners.