According to a study from 2022 published in the National Library of Medicine, 90% of drug candidates fail beyond phase 1 testing. For companies in the business of developing life-saving therapeutics, the process is fraught with risks, such as being very slow and costing billions of dollars. Google is hoping to ease this process with its latest AI model, TxGemma.
TxGemma is an open successor of Tx-LLM, which Google introduced last October to help with drug development. Tx-LLM garnered significant interest from interested parties, such as drug researchers, who wanted a model to use and fine-tune to develop therapeutic applications. To answer this call, Google has released TxGemma for developers who can use this model on their own data and for their specific tasks.
TxGemma is built upon Google’s Gemma models but differs in that it’s trained to help with the development of therapeutics. TxGemma can understand and predict the properties of therapeutics throughout the development process. It can help researchers identify the most promising targets, and it can predict the outcomes of clinical trials so that less time is wasted.
The new TxGemma models are designed to be versatile. The models come in three sizes so that developers can choose the right models depending on the capabilities of the hardware available to them. They are available with 2 billion parameters, 9 billion parameters, and 27 billion parameters. Each features a ‘predict’ version that can handle narrow tasks like classification (e.g., will this molecule cross the blood-brain barrier?), regression (e.g., predicting a drug's binding affinity), and generation (e.g., given the product of some reaction, generate the reactant set).
Demonstrating the effectiveness of its largest 27B parameter model, Google stated:
“The largest TxGemma model (27B predict version) delivers strong performance. It's not only better than, or roughly equal to, our previous state-of-the-art generalist model (Tx-LLM) on almost every task, but it also rivals or beats many models that are specifically designed for single tasks. Specifically, it outperforms or has comparable performance to our previous model on 64 of 66 tasks (beating it on 45), and does the same against specialized models on 50 of the tasks (beating them on 26).”
In addition to the TxGemma models, Google has also released TxGemma-Chat models in 9B and 27B parameter configurations. What’s special about these is that researchers can ask the model to share its reasoning for its conclusions and answer complex questions. These conversations could help researchers speed up the development of therapeutics dramatically.
Google also announced Agentic-Tx based on Gemini 2.0 Pro. Agentic-Tx aims to overcome limitations around up-to-date external information and multi-step reasoning. It’s equipped with 18 tools to help researchers achieve goals, 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
To begin using TxGemma, just head over to Vertex AI Model Garden or Hugging Face. Google said that its models are open, so it expects the community to further iterate on them and publish any improvements that are made. This will enable therapeutics to be developed even faster to save more lives.
)
0 Comments - Add comment