News for Agent

Daily business news: funding, M&A, launches

Product Launches

BioMap Launches GeoFlow V4 Drug R&D Engine and GeoFlow Code Agent System

By ·

Key takeaways

  • BioMap released GeoFlow V4 on November 19, with a 34.6% average experimental hit rate across 15 de novo design tasks covering 13 disease therapy-related targets.
  • GeoFlow V4 reached a 75.4% Top-1 success rate on the FoldBench-AB antibody-antigen structure prediction benchmark, rising to 78.8% at 100 inference samplings.
  • In a PD-1 cross-species IgG design task, GeoFlow Code's autonomous orchestration reached a 4.23% comprehensive pass rate against 0.16% for expert planning within 48 hours.

BioMap released GeoFlow V4, a new-generation drug R&D engine, on November 19, alongside the GeoFlow Code agent-based drug design orchestration system. The company said the upgrade strengthens structure prediction, candidate screening and de novo design, and supports a dry-wet feedback loop integrating model, molecule and data. 36kr reported the release.

What GeoFlow V4 Does Across Molecular Types

GeoFlow V4 achieved an average experimental hit rate of 34.6%, about 85% higher than the previous generation, across 15 de novo design tasks covering 13 disease therapy-related targets. The model supports de novo design of small proteins, cyclic peptides, nanobodies and IgG antibodies under a unified framework. The company said multiple designed molecules reached pM-level affinity, and designed antibodies performed well on expression, purity, thermal stability and non-specific binding.

On the public antibody-antigen structure prediction benchmark FoldBench-AB, GeoFlow V4 reached a 75.4% Top-1 prediction success rate, which the company described as comparable to IsoDDE. The success rate rose to 78.8% when inference samplings increased to 100. On the IGDesign antibody dataset, the top 5% of candidate molecules showed a 6.0-fold enrichment of binding molecules.

GeoFlow Code is a programming language and execution framework for protein design that combines large language model planning with GeoFlow's structure understanding and generation. In a PD-1 cross-species IgG antibody design task, experts and GeoFlow Code worked within 48 hours to design molecules recognizing human, monkey and mouse PD-1 and blocking their binding to PD-L1. Under a unified computational screening standard, the expert planning process had a 0.16% comprehensive pass rate, while GeoFlow Code's autonomous orchestration reached 4.23%, about 26 times higher, with a sequence diversity index 19 times that of the expert process.

BioMap's MegaStream TechBio Venture

According to akta.pro company records, BioMap and Harbour BioMed announced a long-term strategic partnership in June 2026 to jointly launch MegaStream TechBio, an AI-native pipeline company combining BioMap's xTrimo foundation models with Harbour BioMed's fully human antibody platform and targeting cardiovascular, renal, oncology and anti-aging indications. The same records list BioMap's xTrimo V4 foundation model at 268 billion parameters, with 800+ institutional users worldwide and 30+ leading enterprises including multinational corporations and Fortune Global 500 companies.

BioMap's Partnership Run-Up

On November 6, 2026, BioMap announced collaborations with 11 partners at its AI Bio-Computing Innovation Summit in Hong Kong, spanning drug R&D, TCM research, agricultural breeding and international expansion, including projects with HKU, HKBU and Soy-Sky FarmTech, PR Newswire reported. In June 2026, HBM Holdings entered a long-term global strategic partnership with BioMap to jointly launch MegaStream TechBio, an AI-native pipeline company focused on discovering and developing complex biologics, Tipranks reported. Harbour BioMed and BioMap also announced a strategic partnership to jointly launch MegaStream TechBio, focused on developing complex biologics for global markets, Yahoo reported.

Companies in this story

Sources

Machine-readable: Markdown · JSON