Overview

• MSI-SEER uses deep Gaussian processes with Monte Carlo dropout to generate a Bayesian Confidence Score that flags high-uncertainty slides for pathologist review.
• The tool achieved state-of-the-art accuracy in predicting microsatellite instability across large, racially diverse gastric and colorectal cancer datasets.
• By combining MSI-H status with stroma-to-tumor ratios from spatial tile-level analysis, MSI-SEER forecasts immune checkpoint inhibitor responsiveness and uncovers microenvironment patterns linked to outcomes.
• Published in npj Digital Medicine on May 19, 2025, the model’s multi-center validation reported on August 5 demonstrates its generalizability and potential for cost-effective clinical deployment.
• Researchers are gearing up for prospective clinical studies and exploring expansion to other cancer types and integration with multimodal clinical data.