Google's DeepMind has unveiled a new artificial intelligence tool, AlphaMissense, capable of predicting the potential harm of millions of genetic mutations. This tool aims to expedite research and enhance the diagnosis of rare disorders. Specifically, AlphaMissense focuses on missense mutations, which involve a single letter alteration in the DNA code. While many of these mutations are benign, some can interfere with protein functionality, leading to diseases ranging from cystic fibrosis and sickle-cell anaemia to cancer.
How AlphaMissense Operates
AlphaMissense, an offshoot of DeepMind's renowned AlphaFold program, was trained using DNA data from humans and closely related primates. It discerns which missense mutations are prevalent (and likely benign) and which are rare (and potentially harmful). The AI system, when presented with a mutation, produces a score indicating the perceived risk of the genetic alteration. As Dr. Jun Cheng from the research team analogized, it's akin to recognizing if a word substitution in a sentence alters its meaning.
Implications and Reception
The introduction of AlphaMissense has been met with optimism, but also caution. While the tool has shown promise in predicting the effects of mutations, experts believe it's an advancement, not necessarily a revolutionary change.
Joseph Marsh, a computational biologist, remarked that while AlphaMissense is currently among the best predictors, its top position might be short-lived given the rapid advancements in the field. Moreover, while computational predictions can aid in diagnosing genetic diseases, they should be used in conjunction with other evidence sources.
DeepMind's Official Statement
DeepMind's official announcement emphasized the significance of understanding the root causes of diseases in human genetics. With the vast number of possible mutations and limited experimental data, determining which mutations could lead to diseases remains a formidable challenge. The AlphaMissense catalogue, developed using the new AI model, has categorized an impressive 89% of all 71 million potential missense variants as either likely pathogenic or likely benign.
This is in stark contrast to the mere 0.1% that have been verified by human experts. By providing these AI predictions, researchers can gain insights into results for thousands of proteins simultaneously, potentially prioritizing resources and hastening more intricate studies.
Recent Auto-Prompting Breakthrough
Earlier this week, Google DeepMind highlighted a new method for AI to give itself automatic prompts. Named Optimization by PROmpting (OPRO), This method utilizes large language models (LLMs) as optimizers where the AI models work by attempting different prompts until they find one that comes closest to solving a particular task. This technique is described in a research paper and automates the trial and error process that a person would typically do by typing.
In other AI medical news today, Elon Musk's Neuralink company has began human trials. The trials, named the PRIME Study (Precise Robotically Implanted Brain-Computer Interface), will span six years and are aimed at testing the company's technology designed to assist individuals with paralysis in controlling devices.
Specifically, the company is seeking participants with quadriplegia resulting from vertical spinal cord injuries or ALS, who are above 22 years of age and have a consistent caregiver. I reported in May that Neuralink had been approved for its first human trials.
Also this week, Microsoft partnered with the Department of Defense on an AI microscope that could help to detect cancer. Dubbed the Augmented Reality Microscope (ARM), this device, while resembling a conventional microscope, integrates advanced computer vision algorithms.
These algorithms guide medical professionals to focus on areas of concern, generating heatmaps that categorize cells as benign or malignant. This real-time visual aid can be projected onto a monitor for a more in-depth analysis.