Google DeepMind Releases AlphaFold Powered AlphaMissense

Since the majority of missense variations found in the human genome are often uncertain, Google DeepMind has introduced AlphaMissense, an adaptation of AlphaFold fine-tuned on human and primate variant population frequency databases to predict missense variant pathogenicity. Missense variants are genetic mutations that change a single nucleotide in a gene, resulting in a different amino acid being incorporated into the protein.

Through the integration of structural context and evolutionary conservation, this model has achieved leading results across a diverse array of genetic and experimental evaluations, all without explicit training on such data. Furthermore, the average pathogenicity score of genes is capable of predicting their cell essentiality, enabling the identification of short essential genes that existing statistical methods struggle to identify. To benefit the scientific community, a comprehensive database of predictions for all conceivable single amino acid substitutions in humans is provided, with 89% of missense variants being classified as either likely benign or likely pathogenic.

The researchers used AlphaMissense to assess all 71m single-letter mutations that could affect human proteins. When they set the program’s precision to 90%, it predicted that 57% of missense mutations were probably harmless and 32% were probably harmful.

How is AlphaMissence different from AlphaFold? 

Both AlphaMissense and AlphaFold use deep learning algorithms to make predictions about proteins, but they have different applications. The model was trained on a large dataset of missense variants and their effects on protein function, using in silico methods to predict the effects of these variants. It was tested on a separate dataset of missense variants and compared its performance to other existing methods.

However, AlphaMissense is designed to predict the effects of missense variants on protein function, while AlphaFold is focused on predicting the 3D structure of proteins. While AlphaMissense was trained on a large dataset of missense variants and their effects on protein function, the former was trained on a large dataset of known protein structures. 

AlphaMissense has potential applications in personalised medicine and drug development, while AlphaFold has potential applications in drug discovery and understanding protein function. 

It outperforms existing “variant effect predictor” software, offering a more efficient means for experts to rapidly identify the mutations responsible for various diseases. Additionally, this program has the potential to detect mutations previously unrecognized in connection to particular disorders, thus assisting medical professionals in prescribing improved treatments.

AlphaFold was the most cited paper in 2022, giving birth to innumerable real-life applications like finding drug for malaria, and Covid-19, delivering gene therapy and much more. So let’s see how AlphaMissense is going to change the life sciences.

Read the full paper here.

Read more: Protein Wars Part 2: It’s OmegaFold vs AlphaFold