DeepMind AI will Lead to New Treatments for Genetic Diseases
Contents hide
  1. 1. Genetic Mutations
  2. 2. AlphaMissense Leveraging AlphaFold’s Insights
  3. 3. A Step Forward, But Not a Quantum Leap
Total Views: 36

Google’s DeepMind has created its groundbreaking protein structure prediction AI to solve the genetic mutations responsible for various diseases. Their latest tool, AlphaMissense, demonstrates the ability to forecast which protein mutations may lead to health conditions accurately. AlphaMissense was built upon the AlphaFold network.

Genetic Mutations

Genetic mutations directly trigger medical conditions. It can alter proteins, causing diseases like cystic fibrosis and sickle-cell anemia. 

However, researchers have encountered only a tiny fraction of these single-letter “missense mutations” among the vast array of possibilities within the human genome. Most of these mutations appear to have no detrimental impact on health. It is challenging for scientists and healthcare providers to assess new missense mutations they encounter. To address these “variants of unknown significance,” researchers have developed numerous computational tools, many of which now incorporate machine learning approaches.

AlphaMissense Leveraging AlphaFold’s Insights

AlphaMissense builds upon the AlphaFold model, which predicts protein structures based on amino acid sequences. It does not directly determine the structural consequences of mutations. 

It capitalizes on AlphaFold’s structural intuition to identify probable disease-causing mutations within proteins. Additionally, it incorporates a neural network inspired by language models like ChatGPT, trained on millions of protein sequences. It allows us to discern plausible from implausible sequences.

The DeepMind team’s network outperforms other computational tools in distinguishing disease-causing variants from benign ones. It also excels in identifying problematic variants in laboratory experiments that simultaneously evaluate the effects of thousands of mutations. 

Using AlphaMissense, the researchers compiled a comprehensive catalog of possible missense mutations in the human genome, revealing that 57% are likely benign, while 32% may cause disease.

A Step Forward, But Not a Quantum Leap

While AlphaMissense represents an advancement in mutation effect prediction, it falls short of the transformative impact of AlphaFold. Experts like Arne Elofsson from the University of Stockholm and Joseph Marsh from the MRC Human Genetics Unit in Edinburgh acknowledge its significance but caution against overhyping its potential longevity as the best predictor. The evolving field of computational biology may yield even more effective tools in the coming years.

Yana Bromberg, a bioinformatician at Emory University in Atlanta, Georgia, underscores the need for rigorous evaluation before real-world application. Entities like the Critical Assessment of Genome Interpretation (CAGI) play a crucial role in benchmarking the performance of prediction methods against unreleased experimental data. 

The importance of thorough evaluation must be balanced, as relying on predictions without proper assessment could have severe consequences in genetic medicine.

Also read;

Similar Posts