Google releases TxGemma AI models for drug development

Google has announced the development of an innovative set of “open” AI models called TxGemma, designed to revolutionize the drug development process. The announcement was made on Tuesday during a healthcare-focused event in New York.

According to the company’s statement, the new AI models, which are planned to launch through the Health AI Developer Foundations program this month, have a unique ability to process both regular text and structures of various “therapeutic units,” including chemical compounds, molecules, and proteins.

“Developing therapeutic drugs from concept to approved use is a lengthy and costly process, so we are collaborating with the broader research community to find new ways to improve the efficiency of this process,” explained Karen DeSalvo, Google’s Chief Health Officer, in a blog post provided to TechCrunch. “Researchers will be able to ask TxGemma questions that help predict important properties of potential new therapies, such as their safety or efficacy.”

However, Google has not yet disclosed details about the licensing of these models, leaving open questions about the possibility of their commercial use, customization, and additional training. TechCrunch has reached out to the company for clarification and is awaiting a response.

Despite optimistic statements from many companies, including Google’s spin-off Isomorphic Labs, about the revolutionary potential of AI in drug development, the reality proves to be more complex. Although certain successes have been achieved, artificial intelligence has not yet become a magical solution for laboratory research.

Recent years have been marked by high-profile failures in clinical trials at companies using AI for drug development, including Exscientia and BenevolentAI. Moreover, the accuracy of leading AI systems in this field, such as Google DeepMind’s AlphaFold 3, demonstrates significant variability in results.