BlueRun Ventures Portfolio in Action | BioMap × Luchen Tech: AI Powers a New Engine for Innovative Drug R&D

Recently, BlueRun Ventures portfolio company BioMap, a biological computing platform, teamed up with Luchen Tech's Colossal-AI team — a high-performance intelligent scheduling platform for large-scale AI models — to tackle the challenge of protein monomer and multimer structure prediction. Together, they built xTrimo Multimer Open Source, one of the most cutting-edge solutions in the industry.

BlueRun Ventures was BioMap's Series A investor and Luchen Tech's angel investor. BlueRun firmly believes in the intersection and convergence of AI, big data, and other technological innovations with vertical industries. This collaboration within the BlueRun family ecosystem marks the real-world deployment of Colossal-AI — a general-purpose deep learning system for the large model era — in the biopharmaceutical sector. It once again demonstrates BlueRun's matrix-style coverage combining technology and industry both horizontally and vertically, fostering a vibrant interactive ecosystem among its portfolio companies.

On this strategic partnership, BlueRun Ventures commented: BlueRun believes that "speed is the ultimate competitive advantage" for companies. Whether in technological or model innovation, being fast is the enduring moat. In the AI and big data era, companies need to organically leverage AI, big data, and science to build digital platforms that enable spiraling, high-speed iteration of core capabilities. BioMap is China's first life sciences company built on AI, big data, and computation. Luchen Tech is an internationally leading open-source software company for large model high-performance runtime. As an early investor in both outstanding startups, BlueRun is delighted to see them jointly build a core capability loop of "speed as the ultimate advantage" for enterprise growth and evolution.

BioMap is dedicated to building world-leading biological computing large models to drive algorithm development for various bio-computing engine tasks, achieving breakthrough innovations in target discovery and drug design. Luchen Tech is committed to unleashing AI productivity by building Colossal-AI, a general-purpose deep learning system for the large model era that efficiently accelerates the deployment of AI large models. In this collaboration, the two parties introduced multiple large model acceleration techniques including data parallelism, model parallelism, pipeline parallelism, and dynamic axial parallelism, significantly reducing the time and economic costs of protein structure prediction model inference, improving the design and deployment efficiency in the field, and making it possible to train even larger models.

xTrimo Multimer Open Source

1-2x Single-Card Performance Improvement

Distributed Inference for Ultra-Long Sequences

BioMap's xTrimo large model system (Cross-model Transformer Representation of Interactome and Multi-Omics) is the world's first cross-modal pre-trained model system for biology, capable of representing multi-level biological problems from monomer proteins and protein-protein interactions to immune cells and immune systems, providing powerful computational biology capabilities for developing immune-modulatory protein drugs.

The xTrimo Multimer Open Source release is one of the core modules of BioMap's large model system, supporting both monomer and multimer protein structure prediction. Compared to existing industry models such as AlphaFold Multimer, the open-source xTrimo Multimer optimizes the computational complexity of long-sequence inference at the底层 level through CUDA optimization and kernel fusion, achieving significant performance improvements in single-card inference. Relative to AlphaFold2 and OpenFold, the xTrimo Multimer Open Source model delivers 1-2x or greater single-card inference performance improvement at equivalent accuracy.

The xTrimo Multimer Open Source model also supports distributed inference for ultra-long sequences. By employing model parallelism, pipeline parallelism, and Dynamic Axial Parallelism, xTrimo Multimer can efficiently distribute computation and partial memory across different devices, addressing the computational and memory challenges posed by ultra-long sequences. At sequence lengths of 2K to 3K, xTrimo Multimer Open Source using multi-card inference achieves up to 8.47x and 11.15x speedup over OpenFold and AlphaFold 2 respectively, and up to 4.45x speedup over Uni-Fold2.0, substantially improving practical usability in industrial drug design scenarios.

xTrimo Multimer Open Source can also support sequence inference up to 4K in length. At this length, OpenFold and AlphaFold 2 are unable to complete inference due to memory limitations, while xTrimo Multimer Open Source can complete 4K sequence inference in approximately 20 minutes, unlocking new possibilities for drug design.

Colossal-AI

Accelerating Deep Learning Applications in Innovative Drug R&D

Luchen Tech's Colossal-AI is a general-purpose deep learning system designed to comprehensively accelerate industrial AI large model applications. Since its open-source release, it has garnered nearly 5,000 GitHub Stars within months and repeatedly topped GitHub Trending. Its solutions have been successfully deployed at well-known companies across biomedicine, autonomous driving, cloud computing, retail, and semiconductors, receiving widespread acclaim.

Protein structure prediction has long been one of the most important topics in structural biology and a crucial means for humanity to understand gene translation and protein function. Protein-protein interactions form the important structural basis for proteins to carry out biological functions. However, due to the multi-level structure and complex interactions of proteins, accurately predicting three-dimensional structures remains highly challenging.

The efficiency gains of the xTrimo Multimer Open Source model stem from both BioMap's large model system construction and its deep collaboration with Luchen Tech's Colossal-AI team on computational efficiency optimization.

On the collaboration, BioMap CEO Wei Liu said: "Luchen Tech founder Professor Yang You and BioMap Chief AI Scientist Professor Le Song had previously collaborated extensively. As BlueRun Ventures portfolio companies, the close relationship between BioMap and Luchen Tech laid a solid foundation for this joint effort to build the world's first cross-modal pre-trained model system for biology. As a large-scale bio-computing platform, BioMap has world-class teams in drug R&D, biological data, and bio-computing large models. Working hand-in-hand with Luchen Tech's top experts in large model acceleration, a critical frontier area, we believe we will continue to produce disruptive, leading results."

Luchen Tech founder Professor Yang You said: "The latest protein monomer and multimer structure prediction solution from the Colossal-AI team's collaboration with BioMap represents important progress in deploying Colossal-AI — a general-purpose deep learning system for the large model era — in the biopharmaceutical domain. Going forward, both parties will continue to strengthen cooperation on bio-computing large models, jointly accelerating the application and deployment of deep learning in innovative drug R&D."

BioMap Chief AI Scientist Le Song said: "From simulating individual protein structures to identifying interactions between different proteins to mapping protein complexes, BioMap is dedicated to decoding and modeling the complex human immune system, developing breakthrough innovative drugs, programming the immune system, and curing multiple immune-related diseases. The release of the xTrimo Multimer Open Source model marks another step toward BioMap's xTrimo multi-modal bio-computing large model system, achieved through our collaboration with Luchen Tech's Colossal-AI team, leveraging their high-performance computing advantages and BioMap's cutting-edge technological accumulation in bio-computing. As we have always strived to do, we are precisely and efficiently exploring the immune patterns of cancer, aging, and other diseases, systematically overcoming the challenges of applying AI in target mining and innovative drug design, making more diseases predictable, controllable, and curable."

As an investor in both companies, BlueRun looks forward to seeing more portfolio companies achieve win-win cooperation, enabling better integration of business scenarios with core technologies, jointly building an industrial ecosystem that delivers both social and economic value.

The project is now freely available open-source on GitHub. Click to view: https://github.com/hpcaitech/ColossalAI/#xTrimoMultimer

---

Further Reading

BlueRun's First Demo Day "Frontier of Everything" Goes Live, Gathering 70+ Companies

Can Drug R&D Say Goodbye to "Finding a Needle in a Haystack"? | BlueRun Biotech Salon Highlights

Robin Li Founds BioMap, BlueRun Ventures Participates in 100-Million-Yuan Series A Round | BlueRun Family

BlueRun Ventures was established in Silicon Valley in 1998. BlueRun Ventures China was founded in 2005 and is a venture capital firm focused on early-stage startups.

Currently, BlueRun Ventures China manages multiple USD and RMB dual-currency funds with over RMB 15 billion in assets under management, making it one of the largest early-stage funds in China. It invests primarily at Pre-A and Series A stages, covering hard tech and innovative interaction, enterprise technology, new consumer, and healthcare. It has cumulatively invested in over 150 startups including Li Auto, Waterdrop, QingCloud, Guazi.com, Qudian, Songguo Mobility, Ganji.com, Energy Monster, Yuntu Semiconductor, Machenike, CloudSaints Intelligence, Anxin Network Shield, and BioMap.

BlueRun Ventures has been ranked #1 on Zero2IPO's "China Top 30 Early-Stage Investment Institutions" and ChinaVenture's "China Best Early-Stage Venture Capital Institutions TOP30," and was named among Preqin's Top 10 Global VC Fund Managers for Sustained High Returns.

Additionally, BlueRun Ventures has repeatedly received honors such as "China's Best Early-Stage Firm of the Year," "China's Top Venture Capital Firm," "Most Entrepreneur-Friendly Early-Stage Firm of the Year," and "Most Influential Early-Stage Firm of the Year" from media organizations including Forbes China, 36Kr, Cyzone, Caixin Media, CBNweekly, and Jiemian.