TxGemma - New Open Model for Drug Development

One of the biggest challenges in drug development is finding lead compounds beyond the clinical phase, as 90% of candidates fail in the initial trial phase. In this context, TxGemma—an open model collection developed on top of Google's DeepMind Gemma and its family of modern, lightweight open models—represents a breakthrough. TxGemma aims to harness the power of large language models to improve the efficiency of therapeutic discovery, from identifying promising targets to predicting clinical trial outcomes.

TxGemma is the Successor to Tx-LLM

Launched last October, Tx-LLM was trained to perform a range of therapeutic tasks in the drug development process. The model generated significant interest, so the developers quickly fine-tuned it based on user feedback, resulting in TxGemma. The model is available in three different sizes—2B, 9B, and 27B—each with a "predict" version optimized specifically for narrow therapeutic tasks, such as predicting the toxicity of a molecule or its ability to cross the blood-brain barrier.

TxGemma is based on millions of practical examples that enable the model to excel in various tasks—classification, regression, and generation. The largest predict version, at 27B, outperforms or at least keeps pace with the previous Tx-LLM model in almost all tasks tested, and even outperforms many models optimized for specific tasks. Based on detailed performance data, TxGemma produced similar or better results in 64 out of 66 tasks and performed better than the previous model in 45 tasks.

Chat Capabilities and Further Fine-Tuning Options

The developers focused not only on raw predictive capabilities but also integrated chat features into the models. This allows the models to answer complex questions, justify their decisions, and provide feedback in multi-step conversations. For example, a researcher can ask why a particular molecule was classified as toxic, and the model can justify its answer by referring to the molecule’s structure.

The release of TxGemma offers not only an end product but also a customizable platform for developers and researchers. With the included Colab notebook, it is easy to fine-tune the model based on your own therapeutic data and tasks—such as predicting adverse events in clinical trials. Additionally, TxGemma can be integrated with the Agentic-Tx system, which includes 18 advanced tools, such as PubMed and Wikipedia search, as well as molecular, gene, and protein tools. This solution helps combine everyday research workflows with the multi-step inference capabilities provided by agent systems.

Availabilty

TxGemma is available in the Vertex AI Model Garden and on the Hugging Face platform, so anyone interested can explore how the system works, try out its inference and fine-tuning features, and experiment with the complex workflows offered by Agentic-Tx. As an open model, TxGemma also offers the possibility for further development, since researchers can tailor it to their specific therapeutic development needs using their own data.

The advent of TxGemma could open a new chapter in drug development, significantly shortening the process from the laboratory to the patient’s bedside and reducing development costs. 

Share this post
China Wants to Buy 100,000 Nvidia Chips – But Where Will the Restricted Technology Come From?
As political and economic tensions between the United States and China continue to rise, global technological competition shows no signs of slowing. Nvidia, one of the world’s most prominent chip manufacturers, is working on new strategies to maintain its presence in China, despite increasing geopolitical constraints. The company is seeking room to maneuver not only in business, but also through diplomacy.
Phase Transition Observed in Language Model Learning
What happens inside the "mind" of artificial intelligence when it learns to understand language? How does it move from simply following the order of words to grasping their meaning? A recently published study offers a theoretical perspective on these internal processes and identifies a transformation that resembles a physical phase transition.
How AI is Helping to Reduce Carbon Emissions in the Cement Industry
One industry alone is responsible for around eight percent of global carbon emissions: cement production. That’s more than the entire aviation sector emits worldwide. As the world increasingly relies on concrete for housing, infrastructure, and industrial facilities, cement manufacturing remains highly energy-intensive and a major source of pollution. A research team at the Paul Scherrer Institute (PSI) in Switzerland is aiming to change this—by using artificial intelligence to develop new, more environmentally friendly cement formulas.
Where is Artificial Intelligence Really Today?
The development of artificial intelligence has produced spectacular and often impressive results in recent years. Systems like ChatGPT can generate natural-sounding language, solve problems, and in many tasks, even surpass human performance. However, a growing number of prominent researchers and technology leaders — including John Carmack and François Chollet — caution that these achievements don’t necessarily indicate that artificial general intelligence (AGI) is just around the corner. Behind the impressive performances, new types of challenges and limitations are emerging that go far beyond raw capability.
Rhino Linux Releases New Version: 2025.3
In the world of Linux distributions, two main approaches dominate: on one side, stable systems that are updated infrequently but offer predictability and security; on the other, rolling-release distributions that provide the latest software at the cost of occasional instability. Rhino Linux aims to bridge this divide by combining the up-to-dateness of rolling releases with the stability offered by Ubuntu as its base.
SEAL: The Harbinger of Self-Taught Artificial Intelligence
For years, the dominant belief was that human instruction—through data, labels, fine-tuning, and carefully designed interventions—was the key to advancing artificial intelligence. Today, however, a new paradigm is taking shape. In a recent breakthrough, researchers at MIT introduced SEAL (Self-Adapting Language Models), a system that allows language models to teach themselves. This is not only a technological milestone—it also raises a fundamental question: what role will humans play in the training of intelligent systems in the future?