AI detects risk of Parkinson’s from retinal images

Hello Nature readers,
Today we discover a ‘Pandora’s box’ of new protein shapes, learn how an AI can predict disease risk from your retina and hear why breast cancer often spreads to the spine.

‘Pandora’s box’ of new protein shapes

Surprising shapes and unexpected connections have been discovered in the dark corners of the protein universe. Researchers mined a database containing more than 200 million structures of known proteins that Google DeepMind’s revolutionary AlphaFold neural network had predicted. One team identified more than two million clusters of similarly shaped proteins; another discovered a flower-like structure in some distantly related proteins. “This work actually opened up a Pandora’s box of projects,” says computational biologist Joana Pereira. Next: deciding which structures to prioritize for further investigation.

AI guesses disease risk from retinal images

Scientists have harnessed the power of artificial intelligence (AI) to diagnose and predict the risk of developing a bevy of conditions, from eye disease to Parkinson’s disease and heart failure — all from images of a person’s retinas. Unlike earlier AI tools trained on retinal images, RETFound was developed using self-supervised learning — a method similar to the one used to train large-language models such as ChatGPT. “The model somehow learns what a retina looks like and what all the features of a retina are,” says study co-author Pearse Keane. RETFound performed well at detecting ocular diseases such as diabetic retinopathy. Its ability to predict the risk of heart attack, stroke and Parkinson’s was limited, but still better than other AI models.

Why breast cancer spreads to the spine

A newly discovered type of stem cell that gives rise to the backbone also helps to drive the frequent metastasis of breast tumours to the spine. Unlike stem cells from other bones, vertebral skeletal stem cells produce a protein that makes them likely to interact with cancer cells. In mouse experiments, deleting the gene for the protein cut the chance that cancer cells would spread to the animals’ spines by about two-thirds.

Japan builds its own version of ChatGPT

Japan’s government and big technology firms are sinking hundreds of millions of dollars into creating Japanese-language versions of ChatGPT. The AI chatbot and similar large language models (LLMs) “excel in English, but often fall short in Japanese due to differences in the alphabet system, limited data and other factors”, says machine-learning researcher Keisuke Sakaguchi. One LLM, which uses the Japanese supercomputer Fugaku, is expected to be released next year.

How to accelerate progress on the SDGs

The world can’t rely on organic change to achieve any of the 17 United Nations Sustainable Development Goals (SDGs), writes a group of sustainability and policy specialists. Lack of headway is largely a direct result of inaction by governments. But scientists need to play a part in speeding up progress by supporting policymakers, they write. The group highlights three areas in which researchers can make a difference: identifying and removing roadblocks, finding cost-effective ways to achieve all the SDGs simultaneously and developing criteria to assess the impact of different interventions.

The oil that is eating everything

Palm oil is consumed more than any other fat on the planet, animal or plant. It also emerged from a legacy of enslaved people’s labour and environmental destruction that persists today. Food writer Bee Wilson reviews two new books on the omni-substance, exploring its history, impact and possible future. How can we replace something that “can fry like lard, bake like butter, melt like chocolate and whip like cream — at a fraction of the cost”? The answer, Wilson suggests, is to look at why we need so much in the first place.