New AI tool classifies missense mutations as pathogenic or benign
Genetic diseases are caused by variations in our DNA. Some of these variations are harmless, but others can lead to serious health problems. Classifying genetic variants is important for understanding which ones are likely to cause disease and for developing new treatments. A new AI tool called AlphaMissense can classify the effects of missense mutations, which are the most common type of genetic variant. AlphaMissense is based on Google DeepMind’s breakthrough model AlphaFold, which predicted structures for nearly all proteins known to science from their amino acid sequences. AlphaMissense can predict the pathogenicity of missense variants altering individual amino acids of proteins.
What are Missense Mutations?
Missense variants are genetic mutations that involve the substitution of a single DNA letter, resulting in the alteration of an amino acid within a protein. To grasp the importance of this, think of DNA as a language – changing one letter can modify a word, ultimately altering the meaning of a sentence. Similarly, a missense mutation changes which amino acid is produced during protein translation, potentially affecting the protein’s function.
While the average person carries over 9,000 missense variants, most are benign and have minimal to no impact. However, some missense variants are pathogenic, significantly disrupting protein function. These mutations play a crucial role in diagnosing rare genetic diseases and investigating complex conditions like type 2 diabetes, which can result from a combination of various genetic changes.
The Challenge of Classifying Missense Variants
Classifying missense variants will increase our understanding of which protein alterations may contribute to disease. Of the more than 4 million missense variants observed in humans, a mere 2% have been annotated as either pathogenic or benign by experts, leaving the majority categorized as ‘variants of unknown significance’ due to limited experimental or clinical data. This is where AlphaMissense steps in.
How AlphaMissense Works
AlphaMissense is built on top of AlphaFold, an AI model that predicted protein structures from amino acid sequences. It focuses on predicting the pathogenicity of missense variants that alter individual amino acids within proteins. Instead of predicting structural changes in proteins, it leverages databases of related protein sequences and the structural context of variants to assign a score between 0 and 1, indicating the likelihood of a variant being pathogenic. Users can set their own thresholds for classifying variants based on their accuracy requirements.
Impressive Performance
AlphaMissense stands out for its state-of-the-art predictions, surpassing other computational methods. When tested on classifying variants from the ClinVar database and predicting results from biological experiments, it consistently demonstrated its accuracy. Its performance was particularly noteworthy in cases where other methods fell short, highlighting its robustness.
AlphaMissense is still under development, but it has the potential to significantly improve the way that genetic diseases are diagnosed and treated. It could be used to develop personalized treatment plans for patients with genetic diseases, and even to prevent genetic diseases from developing in the first place.
Further Read
https://www.science.org/stoken/author-tokens/ST-1429/full
GitHub: https://github.com/google-deepmind/alphamissense
