Google Unveils TxGemma for Enhanced Therapeutic Development

Google has introduced TxGemma, a new collection of open models designed to enhance the efficiency of therapeutic development by utilizing large language models. Announced on their website, TxGemma builds on Google DeepMind's Gemma models and is specifically trained to understand and predict the properties of therapeutic entities throughout the discovery process.

TxGemma models are available in three sizes—2B, 9B, and 27B—and include both 'predict' and 'chat' versions. The 'predict' versions are tailored for specific tasks such as predicting molecular toxicity, while the 'chat' versions are equipped for conversational data analysis, allowing researchers to engage in multi-turn discussions about therapeutic predictions.

The release also includes a fine-tuning example Colab notebook, enabling developers to adapt TxGemma to their own therapeutic data and tasks. Additionally, Google has developed Agentic-Tx, a therapeutics-focused agentic system powered by Gemini 2.0 Pro, which integrates TxGemma for multi-step reasoning and advanced research workflows.

TxGemma is accessible on both Vertex AI Model Garden and Hugging Face, inviting researchers to explore and fine-tune the models for specific therapeutic development use-cases.