Written By: Lavanya Chavhan B.Pharm
Reviewed By: Vikas Londhe M.Pharm (Pharmacology)
Google DeepMind has launched TxGEMMA in the Google’s Health-Check Up event in New York concluded on 18 March 2025. TxGEMMA is a groundbreaking large language model (LLM) designed to transform conventional drug discovery and development. It is built on the success of TxLLM and forms the foundation for its new agent-based platform, AgenticX. TxGEMMA has the potential to significantly reduce the costs and timelines of the research and approval of the latest medicine.
What Is TxGEMMA?
Therapeutics GEMMA abbreviated as TxGEMMA is a highly specialized language model trained for a strong focus on biomedical data, specifically drug discovery. TxGEMMA is not general LLMs but is tailored to understand the gradation of biological pathways, molecular structures, clinical trial data, and pharmaceutical development processes.
Features of TxGEMMA includes
Fine-tuning for biomedical tasks: TxGEMMA learns from a wide range of high-quality information that is important for discovering new drugs. This includes both publicly available data and private, carefully selected sources. For example, it studies scientific articles from PubMed, which is a large database of medical research papers. It also uses chemical databases that contain information about different molecules and how they behave. In addition, it looks at clinical trial registries, which track the progress and results of medical studies on new treatments, and biomedical patents, which describe new inventions in medicine. By learning from all these trusted sources, TxGEMMA gains a deep understanding of the science behind drug development.
Open-weight accessibility: One of the special things about TxGEMMA is that its creators plan to make its weights (key numbers, the model learns during training) available to the public. These weights are what allow the model to understand and make decisions based on the data it has studied. By sharing them openly, anyone can use, study, and even improve the model. This is different from many other AI models, especially those made by private companies, where the weights are kept secret and only the company can use them. Google’s decision to release TxGEMMA’s weights publically supports and encourages more people from universities, hospitals, and pharmaceutical companies to work together. This can speed up scientific progress and lead to better treatments for patients.
Multi-modal capabilities: In the future, TxGEMMA is expected to become even more advanced by including different types of biological data, not just written or textual information. This means it will be able to work with things like molecular images pictures of molecules at the microscopic level and genomic sequences, which are the complete sets of DNA instructions in living organisms. By combining these various types of data, TxGEMMA will become a multi-modal model, meaning it can understand and learn from many different kinds of biological information at the same time. This will make it much more powerful and effective in discovering new treatments and therapies for diseases.
Conversational AI for Deeper Drug Discovery Insights:
In addition to making predictions, TxGEMMA also comes in special versions designed for conversation called the 9B and 27B chat models. These versions have been instruction-tuned, which means they have been trained to understand and respond to detailed questions and commands, similar to having a knowledgeable research assistant you can talk to. With these chat models; scientists can have in-depth conversations with the AI. For example, they can ask complicated questions about biology or drug development, get clear explanations for why the model thinks a certain molecule might be harmful or helpful, and even carry on an exchange discussion to explore an idea more deeply. This makes the research process more transparent and interactive, helping scientists better understand the model’s reasoning and use its insights more confidently in their work.
TxGEMMA: A Successor of TxLLM
Before TxGEMMA, DeepMind released TxLLM in October 2024, an early experiment focused on translating language modeling capabilities to drug discovery applications. TxLLM proved that LLMs could expressively suggest new molecular targets, predict drug interactions, and assist in clinical trial design. However, TxLLM had some limitations like;
Limited domain-specific optimization
Closed or restricted access for external researchers
Performance bottlenecks when dealing with multi-step drug development workflows
TxGEMMA answered and addressed all these issues with more extensive, focused biomedical training, open weights, and it has ability to be integrated into larger agentic systems like Agentic-Tx.
Agentic-Tx
TxGEMMA is not a standalone model; it is also a part of Agentic-Tx, Google’s new agent-based framework for biomedical research. Agentic-Tx enables multiple AI agents each fine-tuned for specific tasks like target identification, compound optimization, and toxicity prediction, to collaborate intelligently and autonomously.
The Agentic-Tx framework is a smart and powerful system built to improve how scientists do biomedical research, especially when it comes to understanding diseases and finding new treatments. It works like an intelligent assistant that uses large language models (LLMs) advanced AI systems that can understand and generate human-like text. But Agentic-Tx goes even further by combining these language models with up-to-date biomedical knowledge and the ability to think through complex problems step by step. This means it can search for the latest biomedical information, analyze it carefully, and then use that knowledge to suggest treatments that are tailored to a specific patient’s needs. This kind of system has the potential to make drug discovery faster and more accurate.
Agentic-Tx is equipped with 18 tools, including:
TxGemma as a tool for multi-step reasoning
General search tools from PubMed, Wikipedia and the web
Specific molecular tools
Gene and protein tools
Agentic-Tx is positioned to:
Shorter discovery timelines: It helps speed up the early stages of research by automatically generating hypotheses and assisting with preclinical testing, saving valuable time.
Lower costs: By making better predictions early on, TxGEMMA reduces the need for repetitive lab experiments, cutting down on expenses.
Greater innovation: The model can identify new drug targets that traditional methods might overlook, opening the door to breakthrough treatments.
Why TxGEMMA Matters
The process of developing new medicines has become extremely expensive and slow for the pharmaceutical industry. A study in 2020 found that, on average, it costs more than $2.6 billion and takes over 10 years to bring just one drug to the market. This long timeline and high cost make it very difficult to discover new treatments. However, using advanced models like TxGEMMA could help solve some of these challenges. For example, TxGEMMA can help scientists quickly find molecules that are most likely to become effective drugs. It can also make better predictions about whether a compound will be safe and actually work in treating a disease. In addition, it can assist in designing smarter clinical trials that are more likely to succeed, reducing wasted time and resources. Because TxGEMMA’s model weights are openly shared, researchers all around the world from universities to small biotech companies can test, improve, and build on it. This openness could make drug discovery more accessible to everyone, not just large, wealthy pharmaceutical companies.
Conclusion
Google’s TxGEMMA, when used alongside AgenticX and based on the earlier advancements of TxLLM, marks a major turning point in how artificial intelligence can support drug development. This combination of powerful tools represents a paradigm shift a big change in the way things are done in the world of biomedical research. As the challenges of high costs, long timelines, and complex data continue to slow down traditional drug discovery, more scientists are turning to AI for help. Models like TxGEMMA offer a new way forward by making the process faster, more efficient, and more accessible. With continued development and global collaboration, these AI tools could lead to quicker discoveries, more effective treatments, and ultimately, a healthier future for everyone
References:
1. Introducing TxGemma: Open models to improve therapeutics development, Shekoofeh Azizi, 25 March 2025 available from https://developers.googleblog.com/en/introducing-txgemma-open-models-improving-therapeutics-development/
2. Eric Wang, Samuel Schmidgall, Paul F. Jaeger et al, TxGemma: Efficient and Agentic LLMs for Therapeutics, TxGEMMA report available from https://storage.googleapis.com/research-media/txgemma/txgemma-report.pdf
3. Tx-LLM: Supporting therapeutic development with large language models, Eric Wang, 09 October 2024, available from https://research.google/blog/tx-llm-supporting-therapeutic-development-with-large-language-models/
4. TxGemma, Health AI developer foundation, available from https://developers.google.com/health-ai-developer-foundations/txgemma#agentic_orchestration
Add a Comment