Breakthrough in RNA drug discovery, first RNA basic model reveals measurement technology at the level of more than 1 billion nucleotides

Editor | KX

Recently, biotechnology company Atomic AI announced the successful development of the first large-scale language model (LLM) that utilizes chemical mapping data. Atomic AI combines advanced machine learning techniques with the latest structural biology to solve the mysteries of RNA drug discovery

Researchers at Atomic AI have created a new platform component that leverages in-house custom wet Laboratory analysis of large-scale chemical mapping data collected. The scientists collected data on millions of RNA sequences and made more than a billion nucleotide-level measurements. Trained on this data, ATOM-1 develops a rich understanding of RNA, which can then be used to optimize the properties of different RNA patterns.

Atomic AI twitter address: https://twitter.com/AtomicAICo/status/1735659469609037974

Atomic AI published an article on December 14 A preprint paper titled "ATOM-1: A basic model of RNA structure and function based on chemical map data" was published on bioRxiv. In this paper, Atomic AI describes in detail their unique ATOM-1™ platform components. This basic model can accurately predict the structure and function of RNA and plays an important role in improving the development of RNA therapeutics

Paper link: https://doi.org/10.1101/2023.12.13.571579

Dr. Manjunath Ramarao, Chief Scientific Officer of Atomic AI Said:

"ATOM-1 is able to predict structural and functional aspects of RNA and key features of RNA patterns, including small molecules, mRNA vaccines, siRNA and circular RNA, to Help design treatments efficiently. Our goal is to create a streamlined drug discovery process to advance our own pipeline and work with partners to help validate their RNA targets and tools to ultimately deliver them to patients quickly and more effectively The treatments needed."

Stephan Eismann, Ph.D., founding scientist and director of machine learning at Atomic AI said:

"By building a large dataset based on RNA nucleotide modifications and next-generation sequencing, the Atomic AI team created the first RNA-based model. We are excited about the broad application of our model to other aspects of RNA research and its use in optimizing RNA-based drugs. We are excited about the potential of various properties, such as stability and translation efficiency of mRNA vaccines or activity and toxicity of siRNA."

Lack of available high-quality RNA data

RNA-based drugs and RNA-targeted drugs are emerging as promising new ways to treat disease. Optimizing these therapeutics requires time-consuming and expensive experimental screening, while rational design requires an accurate understanding of RNA structure and function. To date, there has been little high-quality RNA data available to the life sciences community. Because existing methods, such as animal models for gathering in vivo information or cryo-electron microscopy (cryo-EM) for determining 3D RNA structure, are difficult to use and time-consuming. Optimizing key RNA therapeutic properties, including stability, toxicity, and translation efficiency, has been challenging due to a lack of “real” data

First RNA-based model trained on chemical map data

To address this design challenge, Atomic AI launched ATOM-1, the first RNA-based model trained on chemical map data, through a data collection strategy developed specifically for machine learning training. Using a small probe neural network on top of ATOM-1 embeddings, we demonstrate that this model has developed a rich internal representation of RNA. Trained with a limited amount of additional data, these small networks achieved state-of-the-art accuracy on key RNA prediction tasks, demonstrating that this approach could enable therapeutic design across the entire RNA field.

Illustration: ATOM-1 probe for secondary structure prediction with high accuracy generalization. (Source: paper)

ATOM-1 is able to predict the secondary and tertiary structures of RNA more accurately than previously published methods

Illustration: A linear probe with 257 parameters trained on a secondary structure generalizes to other RNAs. (Source: Paper)

#Illustration: ATOM-1 probe for tertiary structure prediction demonstrates state-of-the-art accuracy. (Source: Paper)

Notably, in a retrospective analysis comparing ATOM-1 to other computational tools for vaccine design, ATOM-1 outperformed all 1,600 other methods of predicting mRNA stability in solution . Based on these results, new underlying models can be adapted with limited data to predict different properties of RNA, not only determining the structure of RNA but also predicting other key features of RNA therapeutics.

Illustration: Structural predictions of ATOM-1 probe compared to baseline without base model embedding. (Source: Paper)

For the past two and a half years, we have been purposefully designing and collecting data to train our underlying models, said Dr. Raphael Townshend, founder and CEO of "Atomic AI." "Through machine learning and generative artificial intelligence, we now have a unique opportunity that ATOM-1 can be tuned to predict RNA structure and function with high accuracy from a small number of initial data points."

倴Science cover, Atomic AI’s proprietary AI-driven 3D RNA structure engine

Atomic AI is an emerging biotechnology company founded in May 2021 and headquartered in the San Francisco Bay Area. The company is focused on leveraging the fusion of machine learning and structural biology to advance RNA drug discovery. They have developed a proprietary platform that leverages fundamental deep learning models to explore and design RNA-targeting small molecules, RNA-based drugs, and RNA tools

Related articles on Atomic AI's technology "Geometric Deep Learning of RNA Structures" "("Geometric deep learning of RNA structure") has been featured on the cover of Science magazine.

Science Cover article link: https://www.science.org/doi/10.1126/science.abe5650

Atomic AI’s PARSE engine is a An artificial intelligence-driven 3D RNA structure engine that generates RNA structure data sets. By combining fundamental models from machine learning with large-scale in-house experiments in the wet lab, the engine is able to reveal functional binders of RNA targets. Its breakthrough technology predicts structuring with unprecedented speed and accuracy. , ligandable RNA motifs, which is a key obstacle in current RNA drug discovery methods.

By combining advanced algorithms and large-scale experimental biology research, novel RNA-targeted and RNA-based drugs can be designed to treat diseases for which there are currently no marketed drugs

Pass Leveraging our database of 3D RNA structures discovered and engineered, Atomic AI plans to advance the development of a pipeline of rationally designed small molecule drug candidates

Atomic AI has raised a total of $42 million in funding across two investment rounds, the latest Series A financing was obtained in January 2023. Atomic AI has raised a total of $42 million in funding across two investment rounds, the latest of which was Series A funding in January 2023

Atomic AI is leading the way in artificial intelligence-enhanced structural biology, It is supported by an interdisciplinary team of machine learning researchers, medicinal chemists, engineers and experimental biologists, as well as strategic scientific advisors and world-class investors. By changing the design of RNA drugs, they successfully treated untreatable diseases

Atomic AI official website: https://atomic.ai/

Reference content: https://www.businesswire. com/news/home/20231215527488/en/

The above is the detailed content of Breakthrough in RNA drug discovery, first RNA basic model reveals measurement technology at the level of more than 1 billion nucleotides. For more information, please follow other related articles on the PHP Chinese website!