Researchers from MIT and Microsoft have developed an artificial intelligence model designed to aid in the early detection of cancer. The AI system, called CleaveNet, creates peptides that are targeted by proteases—enzymes that are typically overactive in cancer cells. These peptides are coated onto nanoparticles, which can act as sensors to detect the presence of cancer-linked proteases anywhere in the body.
If these proteases are present, the sensors release a signal that can be identified through a urine test. This method could potentially allow for at-home testing and help diagnose specific types of cancer based on which proteases are detected.
"We're focused on ultra-sensitive detection in diseases like the early stages of cancer, when the tumor burden is small, or early on in recurrence after surgery," said Sangeeta Bhatia, John and Dorothy Wilson Professor of Health Sciences and Technology and Electrical Engineering and Computer Science at MIT.
Bhatia and Ava Amini, principal researcher at Microsoft Research and former graduate student in Bhatia's lab, led the study published in Nature Communications. Carmen Martin-Alonso, founding scientist at Amplifyer Bio, and Sarah Alamdari from Microsoft Research were lead authors.
Previous research by Bhatia’s team demonstrated diagnostic sensors for lung, ovarian, and colon cancers using trial-and-error methods to identify suitable peptides. However, many identified peptides could be cleaved by more than one protease, making it difficult to attribute signals to specific enzymes. Despite this limitation, arrays of different peptides produced distinctive sensor signatures diagnostic for various cancers in animal models.
The new approach with CleaveNet uses AI to design peptide sequences tailored for specific proteases. Users can specify design criteria so CleaveNet generates candidate peptides likely to meet those requirements. "If we know that a particular protease is really key to a certain cancer, and we can optimize the sensor to be highly sensitive and specific to that protease, then that gives us a great diagnostic signal," Amini said. "We can leverage the power of computation to try to specifically optimize for these efficiency and selectivity metrics."
Given about 10 trillion possible combinations for a peptide containing 10 amino acids, AI allows researchers to predict useful sequences much faster than traditional methods while reducing experimental costs.
To develop CleaveNet, researchers trained a protein language model using public data on around 20,000 peptides interacting with matrix metalloproteinases (MMPs). The model predicts which peptide sequences will be cleaved by target proteases efficiently.
In their demonstration focusing on MMP13—a protease involved in metastasis—the system generated novel peptide sequences not seen during training but still effective as intended. "When we set the model up to generate sequences that would be efficient and selective for MMP13, it actually came up with peptides that had never been observed in training, and yet these novel sequences did turn out to be both efficient and selective," Martin-Alonso said.
This selectivity may reduce the number of required peptides for diagnosis or therapeutic purposes while aiding biomarker discovery.
Bhatia's lab participates in an ARPA-H funded project aiming for an at-home diagnostic kit capable of detecting up to 30 types of cancer by measuring multiple kinds of enzyme activity beyond just MMPs.
Additionally, CleaveNet-designed peptides could play roles in targeted therapies by ensuring medicines are released only when exposed to relevant tumor-associated enzymes—potentially improving treatment efficacy while minimizing side effects.
The research received funding from La Caixa Foundation, Ludwig Center at MIT, and Marble Center for Cancer Nanomedicine.
