Google Deepmind has developed a brand-new AI model to categorize missense variations. 

The researchers reveal in a study published in "Science" that it classified 89% of all 71 million potential missense variations as either likely harmful or likely benign.

Most missense mutations found in the human genome are not known to have any clinical effects. The researchers show AlphaMissense, an adaptation of AlphaFold that has been fine-tuned using databases of the frequency of variants in human and primate populations to identify the pathogenicity of missense variants. Their model gets the best results on various genetic and experimental standards by combining structural context and evolutionary conservation, even though it needs to be trained on those kinds of data. 

The average pathogenicity score of genes is also a good predictor of their importance to cells. It can find short vital genes that other statistical methods can't find. As a service to the community, the researchers have made a database of estimates for all possible single amino acid substitutions in humans. They have also classified 89% of missense variants as likely benign or pathogenic.

AI tools that can correctly predict how variants will affect traits could speed up research in molecular biology, clinical genetics, and statistical genetics. Experiments to find mutations that cause disease are expensive and time-consuming because each protein is different, and each experiment has to be developed from scratch, which can take months. Using AI predictions, researchers can get a sample of results for thousands of proteins at once. It can help them decide where to put their resources and speed up more complicated studies. 

Missense variant

A missense variant is a mutation in the genetic code that causes a protein to have a different amino acid. DNA can be thought of as a language where the meaning of a word or phrase can be altered by changing only one letter. The translation of a protein can be affected when an amino acid is altered due to a substitution. 

The average person carries more than 9,000 missense variations. While most are harmless and have negligible effects, some are pathogenic and can adversely alter protein function. Rare genetic illnesses can be diagnosed using missense variants because even one missense variant might cause severe health problems. Furthermore, they help investigate complicated disorders like type 2 diabetes, which may result from several genetic alterations.

AlphaMissense

AlphaMissense is based on their ground-breaking model AlphaFold, which predicted the structures of nearly all proteins known to science based on their amino acid sequences. Their modified approach can forecast the pathogenicity of missense variations that change specific amino acids in proteins.

AlphaFold was fine-tuned on labels differentiating variants seen in human and closely comparable ape populations to train AlphaMissense. Variants that are frequently observed are considered benign, while those that are never observed are deemed harmful. AlphaMissense does not predict protein structural changes or other effects on protein stability. Instead, it uses databases of related protein sequences and variation structural context to generate a score between 0 and 1 that roughly rates a variant's pathogenic risk. Users can utilize the continuous score to select a threshold for identifying variations as pathogenic or benign that meet their accuracy criteria.

Evaluation

AlphaMissense can produce state-of-the-art predictions on various benchmarks without training on genetic or experimental data. Their technology outperformed previous computational methods when applied to variants from ClinVar, a public repository of data on the association between human mutations and disease. Their model is compatible with various measures of pathogenicity, and it was also the most accurate strategy for predicting laboratory results.

Conclusion

AlphaMissense expands on AlphaFold to help the world comprehend proteins. One year ago, the researchers published 200 million protein structures predicted by AlphaFold, assisting millions of scientists worldwide in accelerating research and paving the door for discoveries. The researchers are excited to see how AlphaMissense will help answer fundamental problems in genetics and biological science.

The researchers have also made AlphaMissense's forecasts accessible to the scientific community.

Image source: Unsplash