In September 2023 Google’s AI organisation, DeepMind, released a catalogue of genetic mutations intended to accelerate disease diagnosis and enable improvements to medical interventions. DeepMind, a team of “scientists, engineers, ethicists, and more” is “committed to solving intelligence” for the advancement of science and benefit of humanity. The new release is a catalogue of missense mutations that provides an insight into the effect of genetic mutations that affect the function of human proteins.
AlphaMissense is DeepMind’s AI model for the classification of missense variants. A recent publication reveals that it categorised 89% of 71 million possible missense variants as likely pathogenic or likely benign. The paper presents AlphaMissense as a combination of existing strategies:
- Training on weak labels from population frequency data, avoiding circularity by not using human annotations
- Incorporating an unsupervised protein language modelling task to learn amino acid distributions conditioned on sequence context
- Incorporating structural context by using an AlphaFold-derived system
AlphaMissense uses input from an amino acid sequence to predict the pathogenicity of “all possible single amino acid changes” at a given position in the sequence.
“Notably, AlphaMissense does not predict the structural changes of the mutated amino acid sequences but instead predicts pathogenicity as scalar values.”
A missense mutation, or variant, is a DNA change that causes different amino acids to be encoded at a particular position in the resulting protein. Some mutations can alter the function of the resulting protein. DeepMind compares DNA to language, commenting that changing one letter can “change a word and alter the meaning of a sentence”. Just so, a substitution in DNA changes which amino acid is “translated”.
Each person carries over 9,000 missense variants, of which most are benign. However, some are pathogenic and can “severely disrupt protein function”. They can be used in the diagnosis of rare genetic diseases and are also important for studying “complex diseases”, which are caused by a “combination” of changes.
Classifying missense variants is an “important step” in understanding the pathway to disease. Of over 4 million missense variants that have been identified in humans, only 2% have been annotated as pathogenic or benign. This comprises around 0.1% of all 71 million possible variants.
What does this mean for health?
DeepMind states that a “key step” in using this research is “collaborating with the scientific community”. In partnership with Genomics England, the team is exploring how the predictions could help study the genetics of rare diseases. Although the predictions are not intended for clinical use and should be “interpreted with other sources of evidence”, the work has potential to improve disease diagnosis and discovery.
“Ultimately, we hope that AlphaMissense, together with other tools, will allow researchers to better understand diseases and develop new life-saving treatments.”
Taking steps forward
BBC News reported that a “leading independent scientist” identifies this work as a “big step forward”. Professor Ewan Birney, deputy director general of the European Molecular Biology Laboratory, believes it will allow researchers to “prioritise where to look” for areas that cause disease. He thinks AI is going to play a greater role in life sciences with time:
“I don’t know where it’s going to end but it’s changing nearly everything we do at the moment.”
Dr Ellen Thomas, deputy chief medical officer at Genomics England, is excited by the “new tool”.
“It will help clinical scientists make sense of genetic data.”
Will these changes affect your work? How might vaccine development benefit from AI support? For more like this don’t forget to subscribe to our weekly newsletters.