百曜科技
Category: AI in Biotech / Drug Discovery
AI Virtual Cell (AIVC) platform company building foundation models to simulate cellular states, predict drug responses, and enable cell engineering for drug discovery and biomanufacturing. 百曜科技 was founded in 2024. The company is led by 李鑫 (Li Xin). Based in Beijing, China. Team size: 11-50. Total funding raised: $8.5M. Latest round: Series A. Key investors include 国新创投基金 (China Reform Holdings/China Xinsheng), 峰瑞资本 (FreeS Fund), 百度风投 (Baidu Ventures), 道彤资本 (Daotong Capital), 云启资本 (Yunqi Partners), 顺禧资本 (Shunxi Capital).
- Founded
- 2024
- Headquarters
- Beijing, China
- Team size
- 11-50
- Total funding
- $8.5M
Value proposition
Building AI virtual cell foundation models that simulate dynamic cell-state transitions, enabling in-silico drug screening, target discovery, and cell therapy design — replacing costly trial-and-error wet lab experiments with predictive digital biology.
Products and solutions
AURA CellOS (12B-parameter LLM-JEPA single-cell world model), Xcompass cell foundation model interaction platform, Cell Design Platform (gene-edited cell line optimization), Physiological/Pathological Simulation Platform (drug safety & efficacy prediction), AI Smart Cell Lab (automated cell culture with AI decision-making), GeneCompass (multi-species cell foundation model, 126M cells), CellGraphCompass (graph-based single-cell model), CellPolaris (cell fate prediction model)
Unique value
First company globally to apply LLM-JEPA (Joint Embedding Predictive Architecture) architecture to single-cell biology — moving beyond static expression pattern matching to dynamic cell-state understanding. AURA CellOS is the largest publicly reported single-cell foundation model (12B parameters, 390.5M human single-cell transcriptomes). Achieves SOTA with 66% performance improvement over best open-source model on perturbation prediction tasks.
Target customer
Pharmaceutical companies (Big Pharma & biotech), CROs, cell therapy developers, academic research institutions, and biomanufacturing firms
Industries served
Drug Discovery & Development, Cell & Gene Therapy, Biomanufacturing, Synthetic Biology, Precision Medicine
Technology advantage
Proprietary dual-view representation learning (Expression View + Perception View); JEPA-based cross-view prediction in latent space; Dense-to-MoE three-stage training strategy preventing catastrophic forgetting; Multi-species pre-training (human + mouse); Knowledge-enhanced models published as cover articles in Cell Research, Advanced Science, NSR; Won global Virtual Cell Challenge (VCC) pre-season #1 worldwide, finals #2 globally / #1 in China
How they differentiate
Only company using JEPA architecture for virtual cells vs. standard Transformer/LLM approaches used by competitors. First to break through the scaling law bottleneck in AIVC. Strongest academic foundation (CAS Institute of Zoology). Largest publicly reported single-cell training dataset (390.5M cells). Dense-to-MoE training strategy enables sustainable scaling. Chinese domestic champion with national-level strategic backing.
Main competitors
Xaira Therapeutics (US, raised $1B+), Noetik (US), Cellular Intelligence (US), Tahoe Therapeutics (US), Arc Institute (non-profit research), Insitro (US), Recursion Pharmaceuticals (US)
Key partnerships
Baidu (strategic investor via Baidu Ventures), partnerships with domestic and international pharmaceutical companies for model deployment, cell therapy pipeline co-development, early target discovery, and clinical sample stratification (specific partners undisclosed)
Notable customers
Has secured collaboration意向 with domestic and international pharmaceutical companies and CROs for model deployment, cell therapy pipeline co-development, early target discovery, and clinical sample stratification (specific customer names not publicly disclosed)
Major milestones
2023-09: GeneCompass published as Cell Research cover article (world's first knowledge-enhanced multi-species cell foundation model), 2024-11: Company founded, 2025: CellGraphCompass published (first graph-based single-cell foundation model), 2025: Won Virtual Cell Challenge (VCC) pre-season global #1, finals global #2 / China #1, 2025-10: Angel round (FreeS Fund led), 2025-11: Baidu strategic investment, 2026-06: Series A round (China Reform Holdings led), 2026-07: AURA CellOS launched (12B-parameter LLM-JEPA world model)
Growth metrics
Founded Nov 2024; raised 2+ funding rounds within 18 months; AURA CellOS launched July 2026 (12B params); GeneCompass published as Cell Research cover (2023); won Virtual Cell Challenge global #1 pre-season, #2 finals (2025)
Market positioning
Leading Chinese AI virtual cell platform company; positioned as a platform-type startup (vs. pure research or pure AI pharma); one of the few companies globally building the full "data-model-experiment" closed loop; backed by national-level capital (China Reform Holdings) indicating strategic importance
Geographic focus
China (primary) with global expansion plans; competing with US-based AIVC companies (Xaira, Noetik, Cellular Intelligence) and European players
Patents and IP
Multiple foundation models published as cover articles in top journals: GeneCompass (Cell Research cover), CellGraphCompass (National Science Review), CellPolaris (Advanced Science). IP portfolio around LLM-JEPA architecture for single-cell modeling, dual-view representation learning, and Dense-to-MoE training methodology.
About 李鑫 (Li Xin)
Professor & Research Group Leader at Institute of Zoology, Chinese Academy of Sciences; MIT Postdoctoral Fellow; PhD in Developmental Biology from Northeast Agricultural University (joint with CAS); 2022 National Overseas Outstanding Young Talent Program recipient
Official website: https://www.bio-compass.ai/