Google DeepMind has developed AlphaMissense, an artificial intelligence tool that classifies the effects of 71 million ‘missense’ mutations. This marks a significant step in understanding genetic mutations and their potential links to diseases.
Understanding ‘Missense’ Mutations
Missense mutations are single-letter substitutions in DNA that cause a different amino acid within a protein to be produced.
Changing just one DNA letter can lead to different versions of proteins, sometimes causing diseases. On average, each person carries over 9,000 of these mutations, with the majority being benign.
However, a small portion can lead to diseases such as cystic fibrosis or sickle-cell anaemia.
AlphaMissense in Action
AlphaMissense is an AI model developed from DeepMind’s AlphaFold, known for predicting protein structures.
The AI tool focuses on analysing DNA sequences to determine which single DNA letter changes, or missense mutations, are likely to cause diseases.
The tool has categorised 89% of possible genetic typos as either benign or pathogenic.
To refine AlphaMissense, DeepMind fine-tuned AlphaFold based on variants observed in human and closely related primate populations.
The system doesn’t predict changes in protein structure upon mutation. Instead, it uses databases of related protein sequences and the structural context of variants to produce a score.
This score, ranging between 0 and 1, estimates the likelihood of a pathogenic variant.
In a series of tests, AlphaMissense demonstrated impressive prediction capabilities.
When pitted against other computational methods, the tool showed superior performance in classifying variants from the ClinVar public archive. This archive provides data about the relationship between human variants and diseases.
The predictions made by AlphaMissense have been made freely available to the scientific community.
DeepMind, in collaboration with EMBL-EBI, is making these predictions more accessible for researchers through the Ensembl Variant Effect Predictor.
Collaboration also with Genomics England has explored how these predictions could assist in studying the genetics of rare diseases.
While the predictions are not intended for direct clinical use, they offer the potential to improve the diagnosis of rare genetic disorders and the discovery of new disease-causing genes.
Although AlphaMissense and similar tools hold promise, it’s important to note that these are predictions. Laboratory experiments will be essential to validate these AI-generated leads.
The goal is that as the algorithm advances and the interpretation of its results becomes more refined, its utility in diagnosing genetic diseases will continue to improve.
