Problem

Metastasis is the primary driver of cancer-related deaths, yet accurately predicting its onset is a major clinical challenge. Current diagnostic tools often fail to detect occult metastatic cells and metastatic precursor cells, resulting in either overdiagnosis, with unnecessary treatments, or missed metastases that delay critical interventions. This underscores an urgent need for innovative biomarkers and more sensitive assessment methods capable of identifying hidden and early-stage metastatic cells.

Solution

Researchers at OSU have developed EmitGCL, a new computational oncology tool designed to predict future metastasis from single-cell RNA sequencing data. It leverages graph contrastive learning to detect subtle differences between cell groups in primary and lymphatic areas, identifying the ones more likely to spread. The model combines AI modeling, computational analysis, and experimental and clinical validation to ensure the accuracy and applicability of its findings. Beyond prediction, EmitGCL detects key biomarkers and transcriptional drivers of metastasis. When tested in a cohort of breast cancer patients with negative lymph nodes, the framework successfully identified metastatic cells undetectable by other tools.

Applications

• Metastasis risk score calculation and early metastatic prediction
• Gene signature comparison: the tool identifies critical genes responsible for metastasis, potentially providing actionable biomarkers and targets for intervention
• Precision medicine: EmitGCL provides valuable insights to tailor medical treatment to the individual characteristics for outcome improvement

Advantages

• Advanced analytical capabilities: EmitGCL detects metastasis at early stages, complementing existing clinical diagnostic methods
• Enhanced accuracy: the tool has higher true positive rates and lower false positive rates as compared to other diagnostic tools
• Improved performance: EmitGCL outperforms existing computational tools across six cancer types (breast, pancreas, nasopharyngeal, papillary thyroid, head and neck, and lung) in identifying occult metastatic cells with higher sensitivity and specificity
• Multi-omics capability: the framework is built to integrate multiple data types, such as transcriptomics, proteomics, and epigenomics, improving its robustness and transferability across various cancer types

Seeking opportunities for co-development, out-licensing or new venture formation