Google DeepMind AI tool assesses DNA mutations for harm potential


Researchers at Google DeepMind have developed an artificial intelligence program designed to accelerate research and enhance the diagnosis of rare disorders. This program is capable of predicting the potential impact of millions of genetic mutations, categorizing them as either benign or disease-causing.

This AI program specializes in making predictions regarding missense mutations, which involve the alteration of a single letter within the DNA code. Although these mutations may appear harmless on the surface, they have the potential to interfere with the normal functioning of proteins, leading to a wide range of diseases.

These diseases encompass conditions such as cystic fibrosis, sickle-cell anemia, various forms of cancer, as well as disorders related to brain development. The program’s ability to assess the impact of these mutations can significantly aid in understanding and addressing these health challenges.

The research team employed AlphaMissense to evaluate a comprehensive set of 71 million single-letter mutations that have the potential to influence human proteins. When configuring the program to operate with a precision level of 90%, it made predictions indicating that approximately 57% of these missense mutations were likely to be benign, while around 32% were likely to be harmful.

However, the program expressed uncertainty regarding the impact of the remaining mutations, suggesting the need for further investigation and analysis. Based on their research outcomes, the scientists have made a significant contribution to the field by launching a freely accessible online catalog of these predictions.

This resource is intended to assist geneticists and clinicians in two key areas: firstly, in their efforts to study how mutations contribute to the development of various diseases, and secondly, in diagnosing patients who present with rare disorders. This catalog serves as a valuable tool for advancing research and improving patient care in the realm of genetics and rare diseases.

An average person carries approximately 9,000 missense mutations scattered across their genome. Remarkably, of the over 4 million missense mutations observed in humans, only a small fraction, specifically 2%, have been definitively categorized as either benign or pathogenic.

In the medical field, doctors already utilize computer programs to forecast which mutations might potentially contribute to disease development. However, due to the inherent inaccuracy of these predictions, they can currently only offer supplementary evidence to support the diagnostic process.

The development of more precise and reliable AI tools, like the one created by Google DeepMind, holds the potential to significantly enhance the accuracy of such predictions and thus provide invaluable assistance to medical professionals in diagnosing and treating patients with genetic disorders.

Moreover, AlphaMissense has the potential to highlight mutations that have not previously been associated with specific disorders, opening up new avenues for research and potentially guiding doctors toward more effective treatment strategies. This breakthrough in AI-driven genomics promises to bring about substantial improvements in disease diagnosis and patient care.

AlphaMissense represents an adaptation of DeepMind’s pioneering AlphaFold program, originally designed for predicting the three-dimensional structures of human proteins based on their chemical compositions.

To train AlphaMissense, the program was provided with extensive datasets comprising DNA information from humans as well as closely related primates. This data allowed AlphaMissense to discern which missense mutations are prevalent, suggesting their likely benign nature, and which are infrequent and potentially detrimental.

Simultaneously, the program underwent a learning process where it became proficient in the “language” of proteins by analyzing millions of protein sequences. This enabled AlphaMissense to develop a comprehensive understanding of what constitutes a “healthy” protein structure.

By combining these two facets of its training, AlphaMissense has emerged as a powerful tool for predicting the effects of genetic mutations on proteins, which has far-reaching implications for genetic research and disease diagnosis.

When the trained AI is fed a mutation, it generates a score to reflect how risky the genetic change appears to be, though it cannot say how the mutation causes any problems.

“This is very similar to human language,” Cheng said. “If we substitute a word in an English sentence, a person familiar with English can immediately see whether the word substitution will change the meaning of the sentence or not.”

Prof Joe Marsh, a computational biologist at Edinburgh University who was not involved in the work, said AlphaMissense had “great potential”.

“We have this issue with computational predictors where everybody says their new method is the best,” he said. “You can’t really trust people, but [the DeepMind researchers] do seem to have done a pretty good job.”

Read more

Recommended For You