Integrating single-cell transcriptomics with Artificial Intelligence reveals pan-cancer biomarkers of brain metastasis

Abstract

Brain metastasis (BrM) represents a devastating complication across various cancer types, posing as a significant contributor to global morbidity and mortality. Hence, identifying robust biomarkers for early detection across various cancer types with a propensity for BrMs and their therapeutic targeting is highly timely and critical. In this study, we leveraged single-cell RNA sequencing (scRNA-seq) data from six cancer types and combined with convolutional neural network (CNN)-based ScaiVision algorithm to identify a pan-cancer BrM signature that achieved remarkable accuracy in distinguishing BrM from primary tumour cells. Further analysis revealed that the BrM signature was not only prognostic but also detectable in bulk RNA-seq data, providing a stratification tool for patients with high or low metastatic potential. Strikingly, this signature was detected at high levels in the tumour educated platelets, showcasing its potential as a minimally invasive tool for metastasis detection. High BrM signature scores were associated with reduced patient survival, particularly in cancers prone to brain metastasis, such as renal and colorectal cancers. Further analysis uncovered VEGF signalling as a central driver of communication networks in high BrM-scored cells. Accordingly, drug repurposing analysis identified Pazopanib as a candidate for targeting highly metastatic cells that disrupts VEGF signalling networks, and potentially impedes brain metastatic progression in multiple cancer types. This study presents a comprehensive pan-cancer BrM signature with clinical implications for early detection and therapeutic intervention in brain metastasis.