Core Insights - The article discusses the emergence of Generative Science driven by GenAI, highlighting the significance of biological foundation models as a key area of focus in the industry [1] - The release of IntelliFold 2 marks a significant upgrade in biological modeling, showcasing its superior performance compared to AlphaFold 3 in critical tasks related to drug development [4][7] Group 1: Biological Foundation Models - Biological foundation models leverage GenAI architectures like Transformer to extract valuable insights from vast amounts of data, revealing the "grammar of life" that is often difficult for humans to perceive [2] - The AlphaFold series by DeepMind is recognized as a groundbreaking achievement, with AlphaFold 3 being a notable industry benchmark, although few models are expected to match its capabilities by the end of 2025 [2] - The proliferation of biological foundation models faces challenges related to open-source accessibility, performance limits, and deployment convenience, necessitating high-performance and high-availability solutions [2] Group 2: IntelliFold 2 Release - IntelliGen AI's IntelliFold 2, released recently, demonstrates significant advancements in efficiency and functionality, surpassing AlphaFold 3 in key performance metrics [4][7] - IntelliFold 2 achieved a success rate of 58.2% in antibody-antigen interactions, outperforming AlphaFold 3's 47.9%, indicating a more robust model for identifying high-potential candidates [13] - In protein-ligand co-folding tasks, IntelliFold 2 achieved a success rate of 67.7%, surpassing AlphaFold 3's 64.9%, which is crucial for small molecule drug design [13] Group 3: Technical Innovations - The core breakthroughs of IntelliFold 2 stem from a rethinking of "information representation capabilities" and "hardware computing characteristics," integrating microscopic interaction rules with AI computational paradigms [11] - The model employs a random atomic-level tokenization approach to enhance its ability to capture fine-grained atomic contact patterns, addressing challenges in traditional modeling methods [14] - IntelliFold 2's architecture allows for a unified model that connects structure prediction with various functional discoveries, enabling researchers to inject hypotheses and constraints for more controlled and precise task development [17] Group 4: Future Prospects - The article emphasizes the ongoing global competition in biological foundation models, with IntelliFold 2's release being a significant milestone that showcases the potential for emerging teams to achieve outstanding results [25] - IntelliGen AI plans to release a De novo design model for Binder and antibodies in mid-2026, aiming to unify prediction and generation processes to accelerate drug development [26] - The future landscape of biological foundation models is expected to become increasingly competitive, with hopes for new innovators to drive the era of intelligent drug design [27]