Deep learning model accurately predicts microsatellite instability in tumors and identifies uncertain cases

One in every three people is expected to develop cancer in their lifetime, making it one of the most pressing health challenges globally. A key factor in determining cancer prognosis and treatment response is the tumor’s microsatellite status—whether it is microsatellite stable (MSS) or microsatellite instability-high (MSI-H). This status reflects the stability of DNA in tumor cells, particularly in regions known as microsatellites, which are short, repetitive sequences of DNA. When these regions accumulate mutations due to defects in DNA repair mechanisms, the tumor is classified as MSI-H, a condition often linked to better responses to immunotherapy and improved survival outcomes in certain cancers. Traditionally, determining microsatellite status involves time-consuming and costly laboratory tests such as polymerase chain reaction (PCR) or immunohistochemistry (IHC). These methods can sometimes yield ambiguous or uncertain results, leading to delays in treatment decisions. Now, researchers have developed a deep learning model capable of predicting microsatellite instability directly from standard whole-exome sequencing (WES) data. The model analyzes patterns in genetic mutations across tumor samples and identifies subtle signatures associated with MSI-H status with high accuracy. Notably, the system is also able to flag cases where the result is uncertain, allowing clinicians to prioritize these for further testing. This innovation could significantly streamline cancer diagnostics, reducing reliance on separate, specialized tests and enabling faster, more accessible assessment of microsatellite status. By integrating this AI-powered tool into clinical workflows, oncologists may improve treatment planning, particularly for patients who could benefit from immune checkpoint inhibitors. The model’s ability to detect uncertainty is especially valuable, as it helps avoid misclassification and supports more informed decision-making. As AI continues to advance in genomics, such tools may become essential in personalizing cancer care and improving outcomes for millions of patients worldwide.