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.
Sharing TxGemma, a new family of open models from Google DeepMind, built on Gemma 2 and specifically designed to accelerate therapeutics development. This aims to tackle the slow, expensive drug discovery process using specialized AI.
TxGemma comes in two flavors and multiple sizes (2B, 9B, 27B):
TxGemma-Predic focuses on tasks like predicting molecule properties (toxicity, binding affinity, etc.) based on Therapeutic Data Commons benchmarks. The 27B version shows impressive performance, rivaling specialized models.
TxGemma-Chat, on the other hand, adds conversational abilities, allowing you to ask why it made a prediction or discuss complex therapeutic data.
Key aspects:
💊 Therapeutics Focus: Tailored for drug discovery and related biological tasks. 🧬 Gemma 2 Base: Leverages the efficient Gemma 2 architecture. 🧪 Fine-Tunable: Includes a Colab notebook showing how to adapt it to specific therapeutic data or tasks.
🤖 Agent Ready: Can be integrated as a tool within larger AI agent systems (demonstrated with Agentic-Tx). ⚖️ Open Weights (Gemma License): Available on Hugging Face and Vertex AI, allowing broad research and development (check license terms).
This will be a powerful and accessible toolkit for researchers and developers in biotech, pharma, and computational biology.