Recently, Google's DeepMind team launched a new open source model TxGemma, which is specially used to improve the research and development efficiency of therapeutic drugs. The model is based on the fine-tuning and development of Gemma model family, and has powerful language understanding, scientific prediction and multi-round dialogue functions. TxGemma can simulate the entire process of drug research and development, covering multiple key links from early screening to clinical trial prediction, which helps improve R&D efficiency and reduce costs. TxGemma provides three model scales (2B, 9B, 27B parameters), which are suitable for different scenarios. Among them, the "predictive" version focuses on task optimization, while the "chat" version supports multiple rounds of conversations, which facilitates explanation of the reasoning process and answering complex questions. The largest 27B parameter version performed well in testing, performed well in professional models, and performed well in multitasking scenarios. As an open source project, TxGemma has been opened globally through Vertex AI and Hugging Face platforms, with Colab notebook examples to reduce the difficulty of using it. In the future, Google plans to continuously improve its model and welcomes community feedback to facilitate iterative upgrades. Industry experts believe that TxGemma may push drug research and development into a new stage of more efficient and collaborative helping to improve human health.
