AI biology models

WIRED reports that artificial intelligence systems, having already made striking progress on mathematical problems, are now being turned toward the broader sciences. The article suggests AI tools that can reason, prove and model are poised to assist research in physics, biology, chemistry and beyond, raising questions about how scientists will work alongside machines.

Chan Zuckerberg Biohub is partnering with the Department of Energy and the National Institutes of Health in a $1.8 billion initiative to fund large-scale biological data collection for training AI models. The effort aims to give researchers access to vast biological datasets, positioning AI as a core tool for biomedical and energy-related science.

STAT reports on a startup working toward 'general biological intelligence', an ambition to build systems that can understand and model biology the way general AI aims to model cognition. The idea is drawing attention for its scope: rather than targeting a single disease or narrow task, the company is positioning its technology as a general-purpose engine for biological discovery.

A newly reported AI model can predict which molecules membrane transporters carry across cell membranes, a longstanding challenge in biology and drug development. The advance could help researchers understand how cells absorb nutrients and medicines, and accelerate the design of drugs that exploit transporter pathways. Coverage so far is limited, with the claim circulating mainly through science news outlets.

An artificial intelligence company known for building large language models has established its own biology laboratory and is running tests there. The lab is being used to explore how AI models can assist with scientific research, particularly in life sciences. The exact experiments under way have not been fully detailed, but the move signals growing investment by AI firms in hands-on scientific experimentation.

A new artificial intelligence tool reportedly integrates fragmented biological data into a reusable gene network, allowing researchers to combine scattered knowledge about genes and their interactions into one coherent model. The announcement is drawing attention in synthetic biology and bioinformatics circles, where such reusable network models could accelerate research and data reuse across laboratories.

The PROBE benchmark has been introduced as a standardized test for evaluating AI systems that annotate protein functions, and it reportedly reveals significant gaps in how well current models perform. The finding suggests that automated protein-function predictions used in biology research may be less reliable than assumed, prompting scrutiny of AI tools in genomics.

Electric Plant has raised an $11.2 million seed round to build Bombadil, a large language model that interprets electrical signals in plant stems. The system relies on custom hardware sensors attached to living plant tissue using stainless steel screws, translating those signals into data the model can analyse. The project is drawing attention for combining plant biology with AI, and for the unusually ambitious idea of understanding what plants are signalling internally.