Google has unveiled TxGemma, a collection of AI models designed to enhance drug discovery and development. The models aim to streamline the process and accelerate the discovery of new treatments. Developing new drugs is slow, costly, and has a high failure rate.
90% of drug candidates do not progress past phase 1 trials. TxGemma uses large language models (LLMs) to predict therapeutic properties across the research pipeline. This includes identifying promising drug targets and assessing clinical trial outcomes.
Jeremy Prasetyo, co-founder and CEO of TRUSTBYTES, said, “AI that explains its own predictions is a game-changer for drug discovery—faster insights mean faster breakthroughs in patient care.”
TxGemma builds on a model introduced last October. Due to high interest from the scientific community, DeepMind has expanded its capabilities. The open-source TxGemma was trained on 7 million examples and comes in three sizes: 2B, 9B, and 27B parameters.
TxGemma accelerates drug discovery process
Specialized Predict versions are tailored for key therapeutic tasks:
– Classification – Predicting if a molecule can cross the blood-brain barrier
– Regression – Estimating drug binding affinity
– Generation – Inferring reactants from chemical reactions
The 27B Predict model matched or beat specialized models on 64 of 66 benchmark test tasks. TxGemma-Chat allows researchers to ask complex questions, get detailed explanations, and have multi-turn discussions.
This helps clarify the reasoning behind predictions, such as explaining why a molecule may be toxic based on its structure. Google DeepMind has released fine-tuning tools so researchers can adjust the model for their own data. Agentic-Tx integrates TxGemma into multi-step research workflows.
Combined with Gemini 2.0 Pro, it uses 18 specialized tools to enhance research capabilities. Agentic-Tx has been tested on various benchmarks. It can assist with complex research tasks that require reasoning across multiple steps.
TxGemma is now available for researchers and developers to experiment with the models, use fine-tuning tools, and provide feedback.
Image Credits: Photo by Mitchell Luo on Unsplash
April Isaacs is a news contributor for DevX.com She is long-term, self-proclaimed nerd. She loves all things tech and computers and still has her first Dreamcast system. It is lovingly named Joni, after Joni Mitchell.
