On September 11, MGI Tech Co., Ltd. (688114.SH), a global full-stack life science tools provider, and BioMap, a global leader in foundational large models for life sciences, announced a strategic partnership. The collaboration is aimed at developing proprietary cellular foundational models tailored to specific clients and research scenarios, underpinning the creation of dedicated high-quality data production infrastructure for virtual cells and virtual perturbations. This union of an AI4S infrastructure leader and a top-tier AI4S model developer targets the industry's most sought-after asset: high-quality data generation capacity.
Per the agreement, the two parties will execute large-scale experiments involving drug, gene, and environmental perturbations, closing the loop between "experiment, model, and validation" in an integrated wet-dry cycle. They will develop virtual cell property agents for predictive tasks such as hepatotoxicity, immunogenicity, and cell morphology, enabling the large-scale production of high-quality AI-Ready cellular data. This data will, in turn, provide continuous experimental feedback to drive iterative model improvements.
From a strategic standpoint, this partnership represents a complementary fusion of scarce capabilities. As a global pioneer in life science foundational models, BioMap has previously secured the championship at the inaugural 2025 World Virtual Cell Challenge, underscoring its leading prowess in cellular perturbation prediction. The company's proprietary foundational model, xTrimo, with 268 billion parameters, has achieved state-of-the-art (SOTA) performance across more than 300 predictive tasks.
Conversely, MGI possesses cutting-edge cross-omics instrumentation, established experimental systems, and robust large-scale data production capabilities. This collaboration will bridge model R&D with high-throughput experimentation, allowing models to steer experiments while data enriches the models, thereby continuously elevating virtual cell prediction accuracy. High-quality experimental data serves as the most critical "fuel" for large models, and the ability to produce such data at scale is precisely the core asset MGI has cultivated over many years.
For MGI, this alliance affirms its full-stack toolchain as the foundational "base" for frontier research. Beyond conducting one-off experimental validations, its cross-omics instruments and experimental systems will now enable continuous validation, facilitating the large-scale accumulation of high-fidelity real-world data to feed model iteration. Virtual cells (AI Virtual Cell) are recognized as one of the most avant-garde frontiers in the global "AI + life sciences" domain, where simulating cellular responses to drugs, genes, and environmental shifts within computational environments holds the promise of reshaping drug discovery paradigms.
As AI-driven pharmaceutical competition intensifies at the data level, those with superior high-quality data production capacity will secure a decisive advantage in model iteration. Looking ahead, both parties intend to standardize and productize these joint capabilities, exploring integrated "instrument + model + agent" offerings and AI CRO (Contract Research Organization) business models. These efforts aim to deliver combined AI prediction, experimental validation, and iterative optimization support across a broader spectrum of drug development and biopharmaceutical scenarios.