MIT builds AI cancer‑detection sensors
- MIT and Microsoft researchers said January 6 they built an artificial-intelligence system that designs molecular cancer sensors for nanoparticle tests read through urine. - The model, called CleaveNet, searches a design space of about 20^10 possible 10-amino-acid peptides to find protease targets linked to tumors. - The work extends MIT’s earlier urine- and inhaled-sensor cancer tests toward at-home screening. (news.mit.edu)
Cancer sensors work like tripwires: they are built to react when tumor-linked enzymes cut them. MIT and Microsoft researchers said January 6 they used artificial intelligence to design those tripwires faster and more precisely. (news.mit.edu) (nature.com) The enzymes are called proteases, and cancer cells often use them to slice through surrounding tissue as tumors grow and spread. MIT’s approach coats nanoparticles with short protein fragments, called peptides, that get cut by those proteases and then leave a readable signal in urine. (news.mit.edu) The new system is called CleaveNet. MIT News said it was developed by researchers led by Sangeeta Bhatia at MIT and Ava Amini at Microsoft Research, with Carmen Martin-Alonso and Sarah Alamdari as lead authors on the Nature Communications paper published January 6, 2026. (news.mit.edu) (nature.com) The bottleneck is scale. The paper says a 10-amino-acid peptide can be arranged in roughly 20^10 combinations, a search space too large to test one by one in the lab. (nature.com) (microsoft.com) CleaveNet is meant to narrow that search. The model generates peptide candidates and predicts which ones will be cut efficiently and selectively by a target protease, including enzymes that are overactive in cancer. (nature.com) (github.com) MIT said those peptides can be attached to nanoparticles that move through the body and release a signal when they encounter cancer-linked proteases. Doctors could then read the output with a urine test, and MIT said the format could eventually support at-home testing. (news.mit.edu) This did not start from scratch in 2026. In April 2023, MIT reported a paper-strip urine test using DNA-barcoded nanoparticle sensors for early cancer diagnosis. (news.mit.edu) In January 2024, the same broader research program produced inhalable lung-cancer sensors that could be delivered by nebulizer or inhaler and read out through urine. MIT said machine learning helped identify a four-sensor combination that accurately detected early-stage lung tumors in mice. (news.mit.edu) The 2026 advance is the design engine, not a new approved screening test. MIT and the Nature Communications paper describe a way to create better peptide sensors for diagnostics and therapeutics, with cancer detection as one of the main targets. (news.mit.edu) (nature.com) Bhatia told MIT News the group is focused on “ultra-sensitive detection” when tumor burden is still small or when cancer returns after surgery. The next step is whether those AI-designed sensors hold up across more diseases and, eventually, in people. (news.mit.edu)
