Shanghai Jiao Tong University released a protein design model that incorporates AI for efficient and precise design of related functions.

On March 22, Professor Hong Liang's team from Shanghai Jiao Tong University released the protein design model Venus. This team combined AI with protein design and modification, establishing the world's largest protein dataset. The model trained on this dataset can accurately and efficiently predict and design protein functions, transforming protein production from "slow trial and error" to "high-efficiency precise design."

This achievement, in conjunction with industry-leading automated equipment, has been industrialized, transforming protein design from the original "complex science" into today's "simple engineering."

The Venus-Pod (Venus-Protein Outsize Dataset) protein sequence dataset established by the Hongliang team contains nearly 9 billion protein sequences and hundreds of millions of functional labels. It is the largest dataset in the world in terms of data scale and the most annotated with functional labels, and it is also four times the volume of the 2.1 billion protein sequences used for training the well-known model in another industry—the American ESM-C model.

This dataset contains 3.62 billion terrestrial microbial protein sequences, 2.64 billion marine microbial protein sequences, 2.43 billion antibody protein sequences, and 60 million viral protein sequences, covering protein sequence information from conventional surface organisms to extreme environment microbes, especially equipped with hundreds of millions of functional tags (such as the temperature, pH, pressure, etc. at which proteins work).

Hong Liang stated that this dataset constitutes a massive "protein reservoir", making it possible for humanity to mine new proteins or biocatalysts, facilitating the rapid development of biomedicine and synthetic biology. Furthermore, AI large models are expected to learn from vast amounts of data and grasp the evolutionary patterns of natural proteins, providing valuable learning materials for AI to design excellent protein products.

In 2024, the Nobel Prize in Chemistry was awarded to the Google DeepMind team, which used AI technology to accurately analyze the relationship between protein sequences and their three-dimensional structures, solving a fundamental problem that had troubled biologists for 50 years.

However, a real issue is that if the amino acid sequence of a protein is slightly altered, even by just a 1% minor change, the overall structure of the protein may not show significant changes, but its function is likely to deteriorate, or even be completely lost.

In other words, to design a successful protein product, one must not only focus on its three-dimensional structure but also be able to successfully predict and design its function.

Therefore, Professor Hongliang's team "takes a different approach" by no longer focusing on the structure of proteins, but instead directly aiming at the ultimate goal of "function prediction," transforming complex protein design into a demand-oriented process that outputs results through minimal experimentation.

"We have trained the Venus (Morning Star) series of models, which, unlike DeepMind's AlphaFold that predicts protein structures, learns the organizational rules of protein sequences in nature and their relationship with function. Its accuracy in predicting the functionality of protein mutations ranks at the top of the industry list," said Hong Liang.

The Venus series models have two core functions: "AI Directed Evolution" and "AI Enzyme Mining".

The so-called "AI-directed evolution" refers to the Venus series models optimizing various performance aspects of a suboptimal protein product, transforming it into a "hexagonal warrior" to meet application requirements.

"AI Enzyme Mining" refers to the Venus series models that, based on its massive dataset of unknown functional proteins, can "select super-powered warriors" to accurately discover proteins with extraordinary functions that meet stringent application requirements, such as extreme heat resistance, extreme acid resistance, extreme alkali resistance, and extreme gastrointestinal digestion resistance.

The extraordinary functions of these proteins have enormous application potential in biotechnology, pharmaceutical research and development, and industrial production, capable of bringing innovation and breakthroughs to related fields.

At the same time, the world's first low-throughput large-volume protein expression, purification, and functional detection automation system, in conjunction with the Venus series models, can continuously complete the expression, purification, and detection tasks of over 100 proteins within 24 hours, improving efficiency nearly tenfold compared to manual labor. This will significantly reduce the investment of human resources, material resources, and time costs in the research and development process, greatly enhancing the efficiency of protein engineering and synthetic biology research. Its purpose is "AI design, automated experiments," freeing researchers from tedious design and experiments; they only need to pose questions, while AI and automation provide solutions, ultimately transforming complex protein science discoveries into a simple "point-and-shoot" process.

Currently, multiple proteins designed using the Venus series model have been successfully industrialized.

Taking the alkaline tolerance modification of the leading domestic growth hormone company, Jin Sai Pharmaceutical, as an example. Enhancing the alkaline tolerance of proteins has always been a highly challenging task. The Hong Liang team used this model combined with a small number of wet lab experiments in a closed-loop iterative verification, achieving a fourfold increase in the alkaline tolerance of ordinary single-domain antibodies in less than a year, saving Jin Sai Pharmaceutical millions of yuan annually. This achievement has been scaled up to multiple batches of 5000 liters, becoming the world's first protein product designed by a large model and produced at scale.

An innovative application of the Venus series model is the modification project for alkaline phosphatase (ALP) from a certain in vitro diagnostic head company. ALP is widely used as a labeling enzyme due to its high stability and sensitivity; the higher its activity, the greater the detection sensitivity, allowing for the detection of extremely low biomarkers. However, enhancing ALP's activity has always been a challenge. The Venus series model has successfully optimized ALP, increasing its molecular activity to three times that of products from leading international companies, bringing great value to ultra-sensitive diagnostic detection (such as myocardial infarction and Alzheimer's disease). Currently, the modified ALP has entered the 200L scale amplification production phase, marking the successful industrial transformation of the Venus series model.

(Source: Jiemian News)

Source: Eastmoney.com

Author: Jiemian News