AI-Driven Sensors Pave Way for Early Cancer Detection

Researchers from MIT and Microsoft have developed sophisticated molecular sensors using artificial intelligence to facilitate early cancer detection. This innovative approach could significantly reduce cancer mortality rates by identifying cancers at stages when they are more treatable. The findings were published in Nature Communications on March 15, 2024.

The team created an AI model that designs peptides, which are short proteins, targeted by proteases, enzymes that are often overactive in cancer cells. These peptides are coated onto nanoparticles that act as sensors, emitting signals when cancer-related proteases are present in the body. This detection method could potentially allow for a simple urine test, making it easier for patients to access early diagnosis from home.

According to Sangeeta Bhatia, the John and Dorothy Wilson Professor of Health Sciences and Technology at MIT, this research concentrates on “ultra-sensitive detection” of diseases such as early-stage cancers. Bhatia, along with Ava Amini, a principal researcher at Microsoft Research, are the senior authors of the study. The lead authors include Carmen Martin-Alonso, a founding scientist at Amplifyer Bio, and Sarah Alamdari, a senior applied scientist at Microsoft Research.

Utilizing Protease Activity for Diagnosis

Over a decade ago, Bhatia’s lab proposed using protease activity as a marker for early cancer detection. The human genome encodes approximately 600 proteases, which can break down proteins, including structural components like collagen. These enzymes are frequently overactive in cancer cells, enabling them to invade surrounding tissues by dismantling the extracellular matrix that normally holds cells in place.

The researchers’ strategy involves coating nanoparticles with peptides that can be cleaved by specific proteases. When these particles are ingested or inhaled and come into contact with cancer-linked proteases, the peptides are cleaved and subsequently excreted in urine, where they can be detected using a simple paper strip test. Bhatia emphasizes the potential of this method to amplify signals that indicate protease activity deep within the body.

Previously, the team employed a trial-and-error approach to identify peptides for detecting specific proteases. While this yielded diagnostic signatures in animal models, the inability to attribute signals to individual proteases limited its effectiveness. The new study introduces an AI system named CleaveNet, which allows researchers to design peptide sequences tailored to cleave efficiently and specifically by the target proteases of interest.

AI Enhances Peptide Design and Diagnostic Precision

CleaveNet enables users to input design criteria, generating candidate peptides that align with those specifications. This advancement enhances the diagnostic capabilities of the sensors by optimizing their sensitivity and specificity toward particular proteases. Amini states, “If we know that a particular protease is really key to a certain cancer, then optimizing the sensor gives us a strong diagnostic signal.”

Given the vast number of possible peptide combinations—approximately 10 trillion for a sequence of ten amino acids—utilizing AI significantly accelerates the process of identifying useful peptide sequences while also reducing experimental costs.

To develop CleaveNet, researchers created a protein language model to predict amino acid sequences of peptides, similar to how large language models predict text sequences. They trained the model using data from around 20,000 peptides and their interactions with matrix metalloproteinases (MMPs). This dual-model approach allows for efficient prediction of peptide sequences that are likely to be cleaved by specific proteases, demonstrating the potential for applications in diagnostics and therapies.

The researchers specifically tested the MMP13 protease, which facilitates cancer cell metastasis. By targeting MMP13, CleaveNet successfully designed peptides exhibiting high selectivity and efficiency for cleavage, leading to promising implications for both diagnostics and therapeutics.

Bhatia’s lab is currently engaged in an ARPA-H funded project aiming to create at-home diagnostic kits capable of detecting and differentiating between 30 different types of cancer at early stages based on protease activity measurements. The research also explores broader applications by potentially integrating peptides designed with CleaveNet into cancer therapeutics to enhance the targeted delivery of treatments.

Ultimately, combining the efforts of this research with ongoing projects could enable the creation of a comprehensive “protease activity atlas.” Such a resource would enhance research into early cancer detection, protease biology, and AI-driven peptide design, paving the way for significant advancements in cancer diagnosis and treatment.

Funding for this research was provided by the La Caixa Foundation, the Ludwig Center at MIT, and the Marble Center for Cancer Nanomedicine.