Researchers from MIT, Harvard, and the Broad Institute have developed an artificial intelligence system capable of predicting the precise location of virtually any protein within a human cell—offering a powerful new tool for biomedical research, drug development, and disease diagnosis.
Protein mislocalization has been linked to major diseases such as cancer, cystic fibrosis, and Alzheimer’s. However, experimentally identifying where the body’s ~70,000 protein variants reside within cells has long been a time-consuming and expensive challenge. The new model, named PUPS (Protein localization prediction Using Protein sequence and cell State), dramatically reduces that burden by leveraging a novel AI approach.
Unlike existing models that often rely on prior lab data or average out results across cell types, PUPS combines a protein language model and an image inpainting system to analyze both the structure of a protein and the condition of individual cells. The model then outputs an image of a single cell with the predicted protein location highlighted, even if the specific protein or cell line has never been studied before.
“You could do these protein-localization experiments on a computer without having to touch any lab bench,” said Yitong Tseo, MIT graduate student and co-lead author of the study published in Nature Methods. “This technique could act like an initial screening of what to test for experimentally.”
The PUPS system was trained using the Human Protein Atlas but goes far beyond the database’s limited coverage, generalizing predictions to new proteins and cell types. Lab validations showed that the model consistently outperformed baseline AI approaches, exhibiting lower prediction error.
Looking ahead, researchers aim to expand PUPS to model protein-protein interactions and predict protein locations within living tissue. The work was funded by multiple agencies, including the NIH, NSF, and the Department of Energy.
Post a Comment