GeoFlow V2: World's First Universal Protein Foundation Model Surpasses AlphaFold3, Marking a New Milestone in AI-Driven Life Sciences
BioGeometry
Large Molecule Drug Developer
'%20fill='%23FFFFFF'%3E%3Crect%20x='249'%20y='126'%20width='1'%20height='1'%3E%3C/rect%3E%3C/g%3E%3C/g%3E%3C/svg%3E)
The AI protein field has just received a stunning upgrade!
Recently,BioGeometryReleasedWorld's First Universal Protein Foundation ModelGeoFlow V2，surpassing AlphaFold3 and RFDiffusion in one fell swoop, achieving comprehensive evolution in structure prediction accuracy, prediction speed, and design rationality.
AsThe World's First Protein"Structure Prediction + De Novo Design" Integrated Generative Large ModelGeoFlow V2 has achieved "one model, infinite possibilities."Capable of simultaneously addressing two critical tasks: protein structure prediction and de novo protein design.
And2024The Heavyweight Work of the Nobel Prize in Chemistry Winner——Google DeepMind'sAlphaFold3and David Baker's teamRFdiffusion，Can only separately complete structural prediction andProteinDesignTask。
Specifically,GeoFlow V2 inAntibodies and Antigens-Antibody ComplexComprehensive in the accuracy and speed of structure predictionLeadingMainstream Models in the Industry, and support for designated epitopes and specified binding modes, Specified Antigen Structure, multi-conformation prediction, and other complex scenarios.
In de novo design,GeneralModelGeoFlow V2 surpasses in generation quality and structural rationality.ByRFDiffusionSpecifically fine-tuned on antibody dataRFAntibody，Truly Achieve了“Quality by Design”Source Innovation。
It can be said that,ThisIndividualWorld's First ModelAnnouncedAI + Life Sciences Entering the Era of Intelligent Creation —— Restructuring the Underlying Paradigm of Macromolecule Design through Generative AI,GeoFlow V2 will accelerate the R&D process of new drugs, vaccines, industrial enzymes, and other products, profoundly transforming the innovation landscape in biomedicine and biomanufacturing.
For example, in the optimization of a key pharmaceutical intermediate synthesis process,GeoFlow V2 OnlyCompleted the high-performance directional transformation of transaminase in 55 days——Enhanced Catalytic Activity 52.3 times, The left-hand selectivity surged from 0% to99.7%, solving the catalytic problem of key chiral precursor amines.
GeoFlow V2The release marks BioGeometry, a generativeAn AI-driven protein design and R&D platform company takes a crucial step toward the application of AGI in life sciences, bringing Chinese solutions to global innovation.
And this journey to the stars and the vast ocean is destined to require walking hand in hand,GeoFlow V2 Officially Launched Today,FacingGlobal research institutions, industry experts, and innovative teams in the fields of biomedicine, synthetic biology, and protein engineeringOpen Limited-Time Free Trial, BioGeometry sincerely invites like-minded individuals to create the future together.
Trial useAnd Technical ReportLink:prot.design
The release of AlphaFold3 has shocked the entire life sciences community and is expected to become a revolutionary tool for drug development.
AndAlphaFoldCompared to 2,AlphaFold 3 will expand the scope of predictionsExtended toProtein,DNA, RNA, Small MoleculesAlmost all biomolecular structures and interactions, bringing about a molecular revolution.
Just recently,The Creators Behind AlphaFold3——Isomorphic Labs Announces $600 Million Financing Completed, becomingThe largest financing in the life sciences sector in 2025 demonstrates the market's anticipation for innovative technologies.
As early as last yearJune, BioGeometryAntibody Design GenerativeAI ModelGeoFlow，YesThe First in the IndustryIn antigen-Antibody Complex Structure Prediction TaskApproachGenerative AI macromolecule design model at the AlphaFold3 level.
And recently releasedGeoFlow V2 has achieved comprehensive evolution in antibody performance metrics, prediction accuracy, and design rationality, surpassing Alph in multiple tasks.aFold3。
In simple terms,The biggest innovation of GeoFlow V2 lies in the creation ofUnified Atomic Diffusion Model, capable of handling simultaneouslyProtein Structure Prediction andProtein design tasks, effectively avoided"Can only predict or can only generate"Limitations.
InIn multiple tasks of AlphaFold3,The most attention is still on proteins.-Ligand Prediction。AbsoluteMost drugsAll are required.Binding to the target protein,Accurately predicting the binding sites and modes of proteins with other molecules will greatly facilitate drug development.
GeoFlow V2 ThenWith powerful capabilities in protein complex structure prediction and modeling, it can accurately predict docking structures under various constraints, widely applicable to antigens.-Antibody, protein-protein, and protein-small molecule systems。
EspeciallyIn low homology antibodies-Antigen Complex Prediction中, the accuracy of traditional homology modeling methods (such as template alignment based on known structures) decreases.。
As for GeoFlow V2 Targeted inLow HomologyAntigen- Performance measurement of antibody complex structure prediction tasksEvaluation shows,GeoFlow The Top-1 success rate of V2 reaches 45.19%.Compared toGeoFlow V1. Increase by 40%,Significantly Exceed Chai-1、Protenix、AFM2.3 And other excellent models in the industry,Demonstrates powerful modeling capabilities for molecular interactions in living organisms。
In addition, inProtein-Small molecule structure prediction task, GeoFlow V2Compared with the current mainstream models, there is alsoFirst-classPerformance.
Figure:GeoFlow V2 inAntigen-Antibody Complexes and Proteins-Small Molecule Complex Structure PredictionResults
Not only asThis,GeoFlow GeoFlow, the lightweight variant of V2 V2-ab performs excellently in antibody structure prediction,Under the premise of higher prediction accuracy,SpeedCompared toAlphaFold Multimer V2.3 FastCompleted150-250 times,Prediction time compressed to the second level,Can support antibody structure prediction tasks at the million level。
Figure:GeoFlow V2-ab in AntibodyStructurePredictionInferenceIn terms of time comparisonAFM2.3 Exhibits a Speed Advantage of 150 to 250 Times
Comprehensive Enhancement of Antibody Design Capabilities
The global antibody drug market size reached 267.7 billion US dollars in 2024.UltraSix antibody drugs surpass the $10 billion mark, becoming globalOne of the most important forms of medication。
Traditional antibody design mostly relies on high-throughput screening technologies in the laboratory, such as immunization or directed evolution, which are not only costly but also inefficient.
AndGeoFlow V2Represented by generativeThe emergence of AI,Paved a Brand-New Path for Biopharmaceutical Discovery, can directlyTarget-basedDe novo generation of novel antibody molecules bringsRevolutionary Opportunities。
InSevenTreatmentTargetFor example,GeoFlow V2 Based onDesignated Epitope Amino Acid Generation1000 antibodies`, the results showed`GeoFlow V2 Successfully on 5 AntigensReproduced the binding mode of the reference molecule（minimum FR RMSD < 5Å）。
Figure:Antibody structures generated based on 7 targets
Moreover,In regard toHER2Target, etc.In the task of antibody design from scratch,GeoFlowAble toGenerate High-Throughput Library Construction, successfully screened outMultipleLow Nanomolar Affinity, DrugabilityGoodAntibody molecules, some of which have completed functional verification in animals.
In simple terms,GeoFlow V2 Brings Efficiency Improvements and a Design Paradigm Shift to Antibody Engineering, Promising Revolutionary Changes for the Discovery of Difficult Drug Targets, Special Epitopes, and Cross-Species Antibodies.
How Was the First All-Purpose Protein Foundation Model Developed?
The Breakthrough Innovation Behind GeoFlow V2 Comes from a Chinese Innovative EnterpriseBioGeometry。
FounderDr. Tang JianLong-term FocusAI and Life Science Intersection: Founded in 2022, the company focuses on the development of a generative AI protein design platform.
Dr. Tang Jian serves as the director of the Algorithm Research Institute at the University of Montreal, Canada (Mila) Lifetime Professor, Chief Scientific Advisor of the CompanyFather of Deep Learning and Turing Award Winner Yoshua BengioServe as.
Currently, the BioGeometry team has gatheredFrom top universities such as Peking University, Tsinghua University, and Fudan UniversityTalent，CoverBackground in Artificial Intelligence, Bioinformatics, and Structural Biology`, multiple`Members have worked at international pharmaceutical companies.And well-known biotechnology companies.
AsOpenAI Continues to Drive GPT Series Upgrade; BioGeometry, with Its Deep Technical Accumulation, Provides a Solid Foundation for the Iteration of the GeoFlow Series Models.
In the latest publicly available technical report,Revealed globallyThe First All-Scenario Atomic-Level Protein Large ModelHow it was achieved.
Unified Atomic Diffusion Model
As early as2021BioGeometry has proactively applied diffusion models to the generation of molecular 3D structures.
GeoFlow V2Achieved for the first timeProtein Structure Prediction and De Novo Design Unified in a Single Framework, which means the modelSupports bidirectional generation，Not only canInputCompleteSequence Generation Structure, Input Complete Structure to Generate Sequence, and can alsoInputPartial Sequence/Structural Generation ComplementarityTheSequence/Structure。
Compared to the pastIndependent model architectures need to be designed for prediction and generation.GeoFlow V2 Breaks the Separation between Structure Prediction and Generation, allowing the modelMore accurately,More general,Stronger。
Protein Generation Preference Optimization
In recent yearsGenerateStyleAIIn Antibody DesignDeveloping extremely rapidly，But the vast majority of models focus onEnhance Antigen-Specific BindingAffinity, with little consideration for biophysical properties.
This also leads to the fact that, in practical applications, although many models can generate a large number of proteins at the touch of a button, onlyA Small Portion of Antibodies/The enzyme truly possesses developability and clinical safety, but subsequent development is restricted.
BioGeometry has realized this pain point,WillExperimental Data and Prior KnowledgeIntegrated intoIn GeoFlow V2,Thereby achieving precise control over the folding and design process.
Currently,GeoFlow V2Accept variousConditionConstraint, includingEpitope Constraint（Residue to Chain）、Contact Constraints（Residue to Residue) andStructureConstraint(SpecifiedStructure Corresponding to a Part of the Sequence), etc.，Comprehensively optimize in terms of design rationality.
It is worth mentioning that BioGeometrySelf-builtHigh-throughput Wet Lab Capabilities，And combine it withThe Excellence of GeoFlowThe combination of predictive and design capabilities, through multiple"Design-Verification"-Redesign"The realization of proteinIterative Optimization。
With the help of this method,GeoFlowOnly inTwo rounds of enzyme evolution, willAThe catalytic activity of the enzyme has increased.65x, the entire process takes only2 months, greatly shortening the enzyme evolution time.
In summary, the model innovation of BioGeometry is not a castle in the air, but truly integratesThe Practical Application of Deep Learning Methods in the Biopharmaceutical Industry, enhancing the druggability of protein therapeutics and effectively reducing wet-lab costs.
GenerativeAI Leads the Next Generation of Macromolecule Design
BioGeometry Contributes Chinese Wisdom
Last year, outstanding scholars in the field of artificial intelligence successively won the Nobel Prize.Award, the highest殿堂 of the scientific community demonstrated through concrete actionsThe Arrival of the AI4Science Era。
Especially, the Nobel Prize in Chemistry has been awarded to the field of computational protein design and protein structure prediction.DeepMind and David Baker's team highlight the huge potential of AI in life sciences with their cutting-edge innovations.
As the cornerstone of life, proteins play a crucial role in living organisms, catalyzing biochemical reactions, constructing cellular structures, and regulating complex functions within the organism. Their structure, function, interactions, and regulatory mechanisms have always been key areas of exploration for scientists.
At present, scientists are usingAI to Explore the Deepest Mysteries of Biology:Whether it isAlphaFold2 Solves the "Protein Folding Problem" That Has Plagued Structural Biology for Over 50 Years，StillRFDiffusion ImplementationSpecificDe Novo Protein Design——VariousProgressIn showcasingAITheForces of Change。
Furthermore, generativeAI, with its unique architecture such as language models and diffusion processes, is more adept at generating novel and practically valuable proteins, opening up broader horizons for protein science research and fundamentally reshaping the technological paradigm and future landscape of biomacromolecule development.
The industrial application of this breakthrough technology will break the limitations of high cost and long cycle in traditional R&D models, providing strong support for industries such as biomedicine and biomanufacturing.Disease Diagnosis, New Drug Development, Industrial Enzyme DesignMore cutting-edge scenarios are expected to witness revolutionary breakthroughs.
As the commanding height of a new round of technological revolution and a new engine for industrial transformation, biomanufacturing has received high attention from countries around the world, including China.At the end of 2023, biomanufacturing was established as a strategic emerging industry, and its strategic position was continuously reinforced in the 2024 and 2025 government work reports.
According to McKinsey statistics, products from biological manufacturing can cover in the futureProducts manufactured through chemical processes, accounting for 70% of the total, may have a direct economic impact of $2-4 trillion annually on the global economy in the next 10-20 years.
In this process,AI-Driven De Novo Protein Design Technology Plays a Key RoleThis technology accelerates the synthesis of chemicals, pharmaceuticals, and functional materials by precisely regulating key parameters such as the catalytic activity, substrate selectivity, and thermal stability of protein components, injecting strong momentum into the trillion-yuan bio-manufacturing market.
As a generativeA Leader in the Field of AI Protein Design, BioGeometry is Committed to Providing Biomanufacturing Solutions for Traditional Manufacturing Enterprises to Achieve Green and Intelligent Transformation. By Building Large Generative Artificial Intelligence (AIGC) Models to Understand the Language of Life, It Reconstructs the Process of Protein Drug Discovery and Design.
Currently, the company has built two fundamental platforms: an AI-driven large-molecule drug design platform and a high-throughput large-molecule drug wet-lab validation platform, achieving a closed-loop integration of dry and wet experiments. These platforms are empowering partners to achieve precise and efficient large-molecule drug design and optimization.
Based onBreakthroughs in cutting-edge technology and industrial practice accumulation in the field of AI protein design, BioGeometry has alreadyMultiple upstream and downstream companies reach strategic cooperation, jointly promoting the implementation of generative AI in the biopharmaceutical and biomanufacturing fields。
In the future, BioGeometry will continue to be driven by technological innovation, focusing on protein design and development, providing higher-quality Chinese solutions to address global challenges, and promoting the biopharmaceutical and synthetic biomanufacturing industries towards a more intelligent and efficient new era.
Written at Last
FromFrom AlphaFold ushering in the revolutionary era of protein structure prediction to RFDiffusion achieving a major breakthrough in de novo protein design, the evolution of AI in the life sciences has never ceased.
Now,The launch of GeoFlow V2 marks a brand-new milestone in this field.——It has, for the first time, achieved the integration of structure prediction and design, featuring higher accuracy, stronger versatility, and characteristics that are closer to real-world application scenarios.
The rise of DeepSeek has allowed the global technology community to witness the innovative strength of Chinese models, marking that Chinese AI power is steadily moving to the center of the world stage — this trend has become an international consensus.
And in generativeIn the new wave of macromolecule design revolution led by AI, Chinese companies like BioGeometry are also becoming the surging "rising force" in the global AI + life science track, taking on a leading role.