Biohub, a company founded by Mark Zuckerberg and Priscilla Chan, announced the launch of a new advanced artificial intelligence model. Essa technology has been carefully trained to study protein biology in depth, aiming to design molecular structures that can be more useful and precise in treatments. The official announcement of the initiative was made directly on the company’s website, marking a significant step in biomedical research.
Proteins are essential components for life, playing crucial roles such as transporting oxygen around the body and fighting viral invaders. Elas are also fundamental in the formation of muscles, the regulation of cells and the strengthening of the immune system, orchestrating a wide range of biological processes. However, even with advances in the prediction of their complex structures, many proteins are still not completely studied and their specific functions remain a mystery to science.
Nova generation from Modelos Evolutivos into Escada (ESM)
The company is introducing a new generation of Modelos Evolutivos in Escada (ESM), an acronym that stands for “Evolutionary Scale Models”. Este innovative system learns from protein sequences through a process that simulates natural evolution. Esse intrinsic knowledge is then used to digitally represent, map, predict and, crucially, design proteins with greater accuracy.
- The ESM system is articulated around three main artifacts, each designed for a specific and complementary function:
- ESMFold2: A state-of-the-art model that excels in predicting the three-dimensional structure of proteins, a cornerstone for designing new proteins with desired characteristics.
- ESM Atlas: A comprehensive map that catalogs 6.8 billion protein sequences, along with 1.1 billion previously predicted structures, creating a vast biological database.
- ESMC: A cutting-edge protein language model, meticulously trained on approximately 2.8 billion protein sequences collected from the entire diversity of life, enhancing your molecular understanding.
Segundo information released by Biohub, this AI model has the ability to substantially assist in the development of proteins that can bind to specific molecular targets. The technology is capable of designing proteins with high binding affinity, a vital characteristic for the success of innovative therapies. The initial results were validated against five highly relevant targets in the areas of oncology and immunology. The company firmly believes this is a transformative moment in the field of protein design, with the potential to revolutionize the way medicines are developed.
ESMC model operating Mecanismo
ESMC represents the latest culmination of an ongoing research program, whose origins date back to 2019. During the Naquele period, the team of scientists managed to develop the first transformer architecture-based language model specifically designed to analyze protein sequences. The intensive training process of this model seeks to internalize the intrinsic fundamental properties that govern the complex biology of proteins.
Essas properties encompass the essential rules that determine how proteins fold into their specific three-dimensional shapes, as well as how they interact with other molecules in the cellular environment. Adicionalmente, the model seeks to understand how these proteins perform their crucial biological functions. The ability to predict and understand these mechanisms is a decisive step towards manipulating and designing new proteins with therapeutic or industrial applications.
Proteins are complex macromolecules, whose basic structure consists of a linear chain of amino acids. Quando these amino acids come together in a particular sequence, they can form a vast and almost unlimited range of structural combinations. The specific order of these amino acids in the chain is the determining factor that guides the folding of the molecule into a unique and highly specific three-dimensional configuration.
Essa particular three-dimensional structure, in turn, is what precisely defines the exact biological function that the protein will exert within an organism. In the initial study stages, the researchers made a relevant discovery: the models developed were capable of learning and processing information that went beyond mere amino acid sequences. Eles demonstrated the ability to encode the structure and biological function of proteins, including properties that had never been explicitly demonstrated or taught to the model during training.
Potencial therapeutic and precision medicine
Após an in-depth learning of the intricate biological patterns of proteins, the artificial intelligence model has demonstrated a remarkable ability to predict the three-dimensional shape of these molecules with high accuracy. Além’s structure, he was able to decipher its essential biological functions and, most innovatively, generate new proteins. Todo this process occurs entirely within the computational environment, drastically speeding up research.
Essas’s advanced capabilities hold significant transformative potential for future applications in medicine. With the assistance of the model, the design of a protein that binds to a specific molecular target with an ideal combination of strength and selectivity becomes a closer reality. Isso implies the possibility of developing greater therapeutic potential, and in a substantially faster way than classical biochemical approaches, which are time-consuming and laborious.
Este technological advancement is of paramount importance, especially when considering the current scenario of creating treatments based on proteins, such as antibodies used against cancer. Atualmente, scientists need to devote considerable time to identifying which specific protein binds to the correct target and, in addition, ensuring that this binding occurs precisely and effectively. Este is the foundation of so-called “precision medicine”, which seeks highly individualized treatments.
However, the traditional process of protein discovery and validation is widely known to be extremely costly and often time-consuming. With Biohub’s artificial intelligence, it is possible to virtually simulate a vast number of proteins in a short period of time. The technology can then predict which ones are most likely to be effective against a specific target, significantly optimizing resources and time for research and development of new medicines.
Validação and trials in oncology and immunology
Para To empirically validate the functionality and effectiveness of the developed system, Biohub researchers selected a set of proteins directly linked to cancer, tumor growth and the complex functioning of the immune system. Foram chose clinically relevant targets such as EGFR, PD-L1 and CTLA-4, which are crucial biomarkers in diverse oncological and immunological pathologies, posing significant challenges.
The artificial intelligence was then instructed to generate tens of thousands of candidate proteins. Esse massive process was completed in approximately two days, with the main objective of testing which of these proteins would demonstrate the best interaction and affinity against the specific targets previously selected. Subsequentemente, the computational system calculated which of the proteins generated would be most stable and would have the greatest probability of developing into a viable and safe treatment.
The detailed results of the studies indicated that the increase in computational power applied to artificial intelligence has resulted in a considerable improvement in the success rate of protein designs. Essa optimization was particularly notable and significant in the case of antibodies that, by traditional approaches, are considered more difficult to bind to their respective targets with the necessary precision. Isso demonstrates the scalability and effectiveness inherent in the computational approach.
Posteriormente, the best performing proteins, which were designed and selected by artificial intelligence, were subjected to rigorous laboratory tests, simulating real biological conditions. Algumas of them proved the effective ability to correctly bind to defined targets. Além also exhibited the desired stability, attesting to their concrete therapeutic potential and their viability as future drug candidates.
Biohub emphasizes that, although diseases follow common biological patterns, a large proportion of them have individual characteristics, requiring personalized approaches. Para certain diseases, such as cancer and rare diseases, the potential for immediate application of this technology is immense and promising. The company demonstrated that the ESM model can design laboratory-validated protein ligands for five clinically relevant targets in a matter of a few days. Este work substantially changes the speed of the initial stage of the drug development process, democratizing access to advanced tools.