AI models for faster drug development
TxGemma is the open model based on Google's Gemma 2, specialized for therapeutics development. Predict molecule properties, analyze data, or chat about results. Available on HF & Vertex AI.
TxGemma includes fine-tuning examples and Colab notebooks demonstrating how to adapt models using proprietary datasets (e.g., TrialBench for adverse event prediction). Researchers can leverage their own data to create task-specific models tailored to unique research needs.
Agentic-Tx orchestrates TxGemma with 18 tools including PubMed search, molecular tools, and gene/protein databases to answer complex chemistry and biology questions. This enables researchers to tackle reasoning-intensive tasks that single-step predictions cannot solve.
TxGemma-Chat versions (9B and 27B) include instruction tuning for conversational AI, allowing researchers to ask why a molecule was predicted toxic and receive structure-based explanations. This builds confidence in model outputs and supports decision-making in drug development workflows.
TxGemma is released as open-source models on Vertex AI Model Garden and Hugging Face at no cost. Researchers can download, deploy, and modify models freely, avoiding licensing costs and maintaining control over their therapeutic AI infrastructure.
TxGemma provides open models (2B, 9B, 27B) trained on 7 million therapeutic examples to predict molecular toxicity, blood-brain barrier crossing, binding affinity, and clinical trial adverse events. This reduces time from lab to bedside and lowers development costs by enabling researchers to filter promising candidates before expensive trials.