Google DeepMind has announced the launch of TxGemma, an open collection of AI models designed to enhance the efficiency of drug discovery and clinical trial predictions. Built on the Gemma model family, TxGemma aims to streamline the drug development process and accelerate the discovery of new treatments.
The development of new therapeutics is a slow, costly process that often faces a high rate of failure—90% of drug candidates do not progress past phase 1 trials. TxGemma seeks to address this challenge by utilizing large language models (LLMs) to enhance the prediction of therapeutic properties across the entire research pipeline. From identifying promising drug targets to assessing clinical trial outcomes, TxGemma equips researchers with advanced tools to streamline and improve the efficiency of drug development.
Jeremy Prasetyo, co-founder & CEO of TRUSTBYTES, highlighted the significance of AI-driven explanations in drug research:
AI that explains its own predictions is a game-changer for drug discovery—faster insights mean faster breakthroughs in patient care.
TxGemma is the successor to Tx-LLM, a model introduced last October for therapeutic research. Due to overwhelming interest from the scientific community, DeepMind has refined and expanded its capabilities, developing TxGemma as an open-source alternative with enhanced performance and scalability.
Trained on 7 million examples, TxGemma comes in three sizes—2B, 9B, and 27B parameters—with specialized Predict versions tailored for critical therapeutic tasks. These include:
- Classification – Predicting whether a molecule can cross the blood-brain barrier.
- Regression – Estimating drug binding affinity.
- Generation – Inferring reactants from chemical reactions.
In benchmark tests, the 27B Predict model outperformed or matched specialized models on 64 of 66 key tasks. More details on the results are available in the published paper.
In addition to its predictive models, TxGemma-Chat offers an interactive AI experience, allowing researchers to pose complex questions, receive detailed explanations, and engage in multi-turn discussions. This capability helps clarify the reasoning behind predictions, such as explaining why a molecule may be toxic based on its structure.
To make TxGemma adaptable to specific research needs, Google DeepMind has released a fine-tuning example Colab notebook, allowing researchers to adjust the model for their own data.
In addition to its predictive models, Google DeepMind is introducing Agentic-Tx, which integrates TxGemma into multi-step research workflows. By combining TxGemma with Gemini 2.0 Pro, Agentic-Tx utilizes 18 specialized tools to enhance research capabilities.
Agentic-Tx has been tested on benchmarks like Humanity’s Last Exam and ChemBench, showing its ability to assist with complex research tasks that require reasoning across multiple steps.
TxGemma is now available on Vertex AI Model Garden and Hugging Face, where researchers and developers can experiment with the models, use fine-tuning tools, and provide feedback.
