Proteins in the wrong cell spot can cause diseases like Alzheimer’s, cystic fibrosis, and cancer. A human cell has about 70,000 proteins. Proteins are tiny molecules that do important jobs in cells. Finding the locations of proteins by hand takes a lot of time and money. Scientists can only test a few proteins in one experiment.
A new method uses artificial intelligence (AI) to find proteins faster. AI models help by studying large datasets of proteins and their locations. The Human Protein Atlas is a big dataset. It shows where 13,000 proteins are in over 40 cell lines. Cell lines are groups of cells grown in a lab. This dataset covers only 0.25 percent of all possible protein and cell line pairs.
A new AI prediction tool
Researchers from MIT, Harvard, and the Broad Institute created a new computer method. This method predicts protein locations in any human cell line. It works even if the protein or cell is new. The method, called PUPS, looks at single cells, not averages of many cells.
This helps find a protein’s spot in a specific cancer cell after treatment. PUPS uses two models together. A protein language model studies the protein’s structure. The structure comes from its amino acids, which are building blocks of proteins. A computer vision model looks at cell images. It uses three stained images to learn about the cell’s state. Stains are dyes that highlight cell parts like the nucleus, microtubules, and endoplasmic reticulum.
The nucleus holds DNA, microtubules support cell shape, and the endoplasmic reticulum makes proteins. PUPS combines this information to predict where the protein is in a cell. It outputs an image with the predicted spot highlighted. This helps researchers diagnose diseases faster. It also helps find drug targets and understand biological processes.
The researchers say PUPS saves months of lab work. The method, published in Nature Methods, could improve future research by predicting protein interactions in living tissue.