Google DeepMind announced the launch of AlphaProteo, an AI system to help biological and health researchers design novel, high-strength proteins that bind to target molecules with accuracy and strength.
AlphaProteo was trained on the Protein Data Bank (PDB) that enables breakthroughs in science and education by providing access and tools for exploration, visualization and analysis of experimentally-determined 3D structures from the PDB archive.
Due to the structure of a target molecule and a set of favorite binding locations on that molecule, AlphaProteo creates a candidate protein that binds to the target.
The tech giant said binders have the potential to open new areas of research in drug development and diagnostic biosensors.
“AlphaProteo can generate new protein binders for diverse target proteins, including VEGF-A, which is associated with cancer and complications from diabetes. This is the first time an AI tool has been able to design a successful protein binder for VEGF-A,” the Protein Design and Wet Lab teams at Google DeepMind said in a blog post.
“AlphaProteo also achieves higher experimental success rates and three to 300 times better binding affinities than the best existing methods on seven target proteins we tested.”
To test AlphaProteo, the AI’s developers designed binders for various target proteins, including “two viral proteins involved in infection, BHRF1 and SARS-CoV-2 spike protein receptor-binding domain, SC2RBD, and five proteins involved in cancer, inflammation and autoimmune diseases, IL-7Rɑ, PD-L1, TrkA, IL-17A and VEGF-A.”
The binding success rate for one viral target, BHRF1, was 88%, on average, ten times higher than traditional methods.
The Google DeepMind web lab team worked with outside research groups, including researchers at the Francis Crick Institute, where data confirmed that AlphaProteo binders prevented SARS-CoV-2 from infecting human cells.
AlphaProteo demonstrated that it could reduce the time required for initial experiments involving protein binders for various uses.
However, despite the breakthroughs, the researchers noted that the AI system has limitations.
For example, AlphaProteo did not generate successful binders for TNFɑ, a protein associated with autoimmune diseases such as rheumatoid arthritis.
“We selected TNFɑ to robustly challenge AlphaProteo, as computational analysis showed that it would be extremely difficult to design binders against. We will continue to improve and expand AlphaProteo’s capabilities with the goal of eventually addressing such challenging targets,” the authors wrote.
The AlphaProteo research team plans to work with the scientific community to observe AlphaProteo’s impact on other biological problems to understand its limitations further.
Additionally, the team has been exploring its drug design use at Isomorphic Labs.
THE LARGER TREND
In June, Google Research and Google DeepMind released a paper announcing the creation of a new LLM for drug discovery and therapeutic development dubbed Tx-LLM, fine-tuned from Med-PaLM 2.
The tech giant’s Med-PaLM 2 is a generative AI technology that uses Google’s LLMs to answer medical questions.
In May, a study performed by Google Research in collaboration with Google DeepMind showed that the tech giant expanded the capabilities of its AI models for Med-Gemini-2D, Med-Gemini-3D and Med-Gemini Polygenic.
Google said it fine-tuned Med-Gemini capabilities using histopathology, dermatology, 2D and 3D radiology, genomic and ophthalmology data.
In 2023, Google released MedLM, two foundational models built off Med-PaLM 2, designed to answer medical questions, generate insights from unstructured data and summarize medical information.
The company said that, through piloting its LLMs with healthcare organizations, it has learned the most effective AI models are designed to address specific use cases.
As a result, the large model of MedLM is made to address complex tasks, while the other is a medium model that can be fine-tuned and scaled across various tasks.
The HIMSS Healthcare Cybersecurity Forum is scheduled to take place October 31-November 1 in Washington, D.C. Learn more and register.
Publisher: Source link
