Google's AI Can Design Cancer Drugs, But It Made a Bizarre Mistake

Описание: Welcome to Summarized Science! Today we're diving into the world of artificial intelligence and medicine. Google DeepMind's AlphaFold 3 is a revolutionary AI that can predict the 3D shapes of molecules, a critical step in designing new drugs. Researchers put it to the test to see if it could help create new antibody treatments for cancer.

Antibodies are a key part of our immune system, and scientists can design them to target specific things, like a protein on cancer cells called CD47. This protein acts like a 'don't eat me' signal, protecting cancer from being destroyed. The goal was to see if AlphaFold 3 could not only predict the structure of these antibody-cancer complexes but also figure out which antibody would be most effective.

The results were a mix of incredible success and baffling failure. While the AI could predict structures with stunning accuracy, it failed a crucial real-world test and revealed a strange new problem that researchers are calling 'reverse docking', where the AI confidently predicted a drug would work by sticking it on completely backwards.

Cited paper:
Y. Xu et al. (2025). Exploring AlphaFold 3 for CD47 Antibody-Antigen Binding Affinity: An Unexpected Discovery of Reverse docking. arXiv:2511.14676v1. http://arxiv.org/abs/2511.14676v1

Images shown are page renders from the paper PDF for commentary/education.