TranX Launches Open AI Platform for Macromolecular Design, Rivaling AlphaFold2

In July, the DeepMind team and the European Bioinformatics Institute announced that they had successfully predicted 214 million protein structures across more than one million species using AlphaFold, covering nearly all known proteins on Earth.

In November, ESMFold, developed by the tech company Meta, successfully predicted the three-dimensional structures of more than 600 million proteins, with its prediction speed reaching up to 60 times that of AlphaFold.

Regarding AlphaFold, Shi Yigong, President of Westlake University, once stated, “This is the most significant contribution artificial intelligence has ever made to the scientific community, and one of the most important scientific breakthroughs achieved by humanity in the 21st century.” AlphaFold has greatly advanced researchers’ exploration and study in the field of biology, expanding the scope of protein function analysis and its downstream applications, and may even transform experimental workflows and outcomes in structural biology.

Thanks to AlphaFold, the field of “AI + biology” has entered a period of explosive growth.As one of the earliest companies in China to deploy AI+protein computing, Tianrang has also ushered in its own breakthrough. In 2021, its independently developed domestic protein structure prediction platform, TRFold, achieved a score of 82.7 (out of 100) in the evaluation of the CASP14 protein test set, second only to AlphaFold2 (91.1 points).

CASP is one of the most authoritative and prestigious competitions in the field of computational biology, hailed as the “Olympics of protein structure prediction.”Tianrang achieved the best performance among all publicly available protein structure prediction models in China at the time, marking that China’s computational biology has entered the world’s first tier.

Dr. Miao Hongjiang is the key figure who led the Tianrang team to achieve this accomplishment.

Miao Hongjiang began his overseas education at an early age. He went to Singapore for middle school. For his undergraduate and master’s degrees, he studied Mathematics and Statistics at the University of Oxford in the United Kingdom, and he pursued his Ph.D. in Computational Biology at Imperial College London.

During his research career, Miao Hongjiang participated in multiple bioinformatics projects involving human genomics, genetic metabolomics, and proteomics. He developed a method for three-dimensional protein structure prediction that integrates the principles of TBM and FM based on target protein contact maps, increasing predictable coverage by more than 20%.

In past biological research, constrained by experimental equipment and technical conditions, researchers were typically limited to using X-ray diffraction for high-resolution structural analysis of proteins, a method that requires the proteins to be crystallizable.

However, protein crystallization itself is an extremely difficult task.

At that time, such cases were common in the industry: even after researchers had obtained clear experimental results regarding a protein’s characteristics, functions, and pathways, they were still unable to accurately determine its structure due to the inability to crystallize the protein. Consequently, they could not explain these results or functions from a structural or mechanistic biological perspective.

“Therefore, we use computational techniques to simulate protein structures, thereby addressing problems that cannot be solved through experimental or physical methods. This was the primary focus of my doctoral research,” explained Miao Hongjiang.

After earning his Ph.D., Miao Hongjiang received a postdoctoral offer from the University of Cambridge.

In 2019, Miao Hongjiang returned to China to apply for a visa, during which time he accidentally met Dr. Xue Guirong, the founder of Tianrang. Dr. Xue is an expert in the fields of artificial intelligence and big data. He previously served as the head of Alibaba’s Mama Big Data Center and Chief Data Scientist, where he was responsible for developing Alibaba’s search engine. In 2016, Dr. Xue founded Tianrang and led his team to achieve significant breakthroughs in AI Go, transportation, finance, and other sectors.

Miao Hongjiang and Xue Guirong first met in early 2019. It was not until late 2018, when AlphaFold1 emerged, that people witnessed for the first time how AI models demonstrated the feasibility of protein structure prediction.“Prior to this, even John Moult, the organizer of CASP, was skeptical about whether he would live to see the problem of protein folding solved.”

By chance, they met for the first time and quickly formed a deep connection despite their brief acquaintance. “On the day we met, we spent the entire afternoon talking,” recalled Miao Hongjiang. That afternoon, their conversation began with an AI Go project and extended to the skepticism expressed by industry veterans regarding the integration of AI and biology.“In 2019, the industry lacked strong confidence and a clear entrepreneurial drive to leverage AI for solving biological problems; however, we all held a long-term vision and a willingness to invest in this endeavor.”

Thus, while others were still hesitating, Tianrang decided to establish the XLab team to pioneer the AI-plus-biotechnology sector. After several in-depth discussions with Xue Guirong, Miao Hongjiang further decided to decline an offer from the University of Cambridge and remain in China to serve as the head of Tianrang’s XLab, embarking on the development of an AI-driven protein simulation platform.

It is no easy feat for a startup to carve out a new domain, nor for a core team to build a new platform.

“Applying AI to the field of protein design and achieving results has only happened in the past two or three years,” explained Miao Hongjiang.

Previously, traditional protein design approaches required meeting numerous conditions. First, researchers needed to leverage long-accumulated understanding and experience of protein structures, combined with complex computational methods, to obtain the designed sequence of the target protein, followed by DNA synthesis and protein expression. Finally, biotechnological techniques were employed to assess the function of the expressed proteins and verify whether they matched the design objectives. This process was not only time-consuming but also incurred substantial costs.

Tianrang has pioneered an entirely new path: AI-driven, on-demand design of proteins with specific functions. This marks another milestone for XLab, following its major breakthroughs in the field of protein structure prediction.

In June this year, Tianrang XLab announced that it had successfully performed de novo design of an iL-2 biosimilar protein using its self-developed protein design platform, TRDesign, to selectively activate the anti-tumor activity of lymphocytes. This signifies that protein drug design capabilities based on TRDesign are no longer constrained by the reliance on known natural proteins in current protein drug development. Instead of passively searching for and discovering therapeutic candidates, researchers can now proactively design therapeutic proteins. Consequently, diseases previously lacking effective treatments may become treatable through the design of potent protein drug candidates.

“Most domestic AI-driven pharmaceutical companies are concentrated on small-molecule drug development. This field only gained significant traction in 2018, when accurate AI-based protein structure prediction was not yet feasible, let alone the R&D of protein-based therapeutics,” Miao Hongjiang told VCBeat New Medicine. He noted that there are actually few companies in China using AI for the simulation or development of large-molecule drugs, with Tianrang being among the first institutions to enter this space.

The capability to design and develop protein-based therapeutics stems from Tianrang’s establishment of an observation system that correlates structure with function. This system creates a high-speed pathway linking sequence, structure, and function, enabling efficient prediction of protein three-dimensional structures from amino acid sequences. Consequently, it allows for the generation of protein sequences and structures that meet target functional requirements through inverse folding.

In addition to its forward-looking strategic layout, Tianrang XLab’s design methodology and philosophy are also epoch-making in significance. In the early stages, researchers favored random design approaches for protein engineering to explore proteins that do not exist in nature. Since the 21st century, scientists represented by Professor David Baker have begun to leverage computational power based on rational design to endow proteins with specific functions and characteristics. However, due to limitations in human understanding of protein structures, the range of achievable functions has remained quite restricted.

Tianrang XLab designs proteins with target functions based on its self-developed AI capabilities, starting from the functional requirements of proteins.By leveraging a suite of end-to-end algorithms for reverse computational analysis, validation, and optimization, we empower drug researchers to design protein lead molecules with specific functions on demand, thereby developing more effective therapeutic approaches.

In just three years, Tianrang XLab has successfully launched TRFold2, a single-chain protein structure prediction platform; TRDesign, a protein design platform; TRComplex, which focuses on complex structure prediction; and TRFold-single, which predicts protein structures without relying on MSA information.

Why Has Tianrang XLab Achieved a Series of Results So Rapidly in the Early Stages of Industry Development? This Is Inseparably Linked to Their Possession of a Universal Underlying Logic.

As a company driven by core AI capabilities, Tianrang possesses a comprehensive suite of sophisticated algorithms and computing power systems. From AI-powered Go to city-level intelligent transportation, Tianrang consistently leverages AI to address challenges in complex systems, with protein structure prediction being no exception. Backed by massive-scale computing clusters and an architecture that seamlessly integrates algorithms with computational power, Tianrang has rapidly entered the field of biological computing.

Furthermore, Tianrang XLab will test and integrate advanced open-source algorithms available on the market, establishing a comprehensive workflow designed to help R&D personnel address practical challenges and genuinely enhance their research and development efficiency.

In October this year, Tianrang XLab officially launched xCREATOR, the first protein design workbench in China, offering free access to domestic research universities and institutions. It aims to provide systematic support for protein researchers, addressing algorithm, data, and computing power needs through a one-stop solution, thereby accelerating the large-scale development and implementation of protein design.

xCREATOR integrates cutting-edge, diverse AI algorithms with robust computational resources to deliver more efficient, convenient, and user-friendly services for protein structure prediction and design. Users can perform tasks such as protein prediction and design without writing any code, and visualize and analyze the computational results. The platform is applicable to peptides, enzymes, antibodies, and various functional proteins.

With the aid of this platform, users can obtain protein structures with near-experimental resolution accuracy within minutes. In the past, this process could take months or even years and required the use of expensive, specialized instrumentation.xCREATOR will greatly help researchers free themselves from tedious engineering and experimental work, allowing them to devote more time to meaningful biological innovation and research.

Unlike the algorithmic "toolbox-style" platforms previously used by R&D personnel, the xCREATOR Workbench places greater emphasis on the design and optimization of protein-related task workflows. It enables researchers to freely categorize, link, integrate, and manage formerly fragmented tasks through a project management approach. On the Workbench, they can complete their R&D work in a one-stop, end-to-end manner—covering project planning, implementation, computation, and analysis—thereby effectively empowering project advancement and management.

Dr. Miao Hongjiang stated, “Many industries in China currently lack core software. Our goal is to provide an ‘EDA software’ for the field of protein design, making it more accessible for biologists.” The xCREATOR platform not only supports individual users in freely conducting various protein-related tasks but also enables team collaboration and sharing of R&D outcomes, thereby addressing practical needs in drug development such as target discovery, protein design, and druggability optimization.

Miao Hongjiang remarked, “We believe that in the foreseeable future, by leveraging AI technology to conduct comprehensive multi-omics analysis at the molecular level of the intracellular microenvironment based on an individual’s global microscopic protein landscape, and with AI-designed protein therapeutics as the core, it will be possible to develop effective treatments for all diseases worldwide.”

Today, Tianrang XLab continues to develop and optimize AI capabilities for biological computing, focusing on protein structure-function relationships, interactions, mutations, and design. Miao Hongjiang revealed,“Tianrang XLab, as an incubated division of Tianrang, has increasingly developed the commercial capability to operate independently. We are advancing financing efforts centered on the XLab project to expand our team and capabilities, thereby accelerating the implementation of the project.”