TMBclaw: tumor clone-aware graph learning improves immunotherapy response prediction across heterogeneous cohorts

Area of Science:

• Oncology
• Computational Biology
• Immunotherapy

Background:

• Immune checkpoint inhibitors (ICIs) are crucial in cancer therapy, but effective patient stratification requires better biomarkers.
• Current tumor mutation burden (TMB) metrics don't fully capture immunogenic neoantigen presentation due to tumor clonal heterogeneity.
• Limited cohort sizes in studies hinder the generalizability of prognostic models.

Purpose of the Study:

• To develop an advanced framework, TMBclaw, for predicting immunotherapy response by integrating clonal dynamics.
• To improve prognostic accuracy and patient stratification in oncology by addressing tumor heterogeneity.
• To enable cross-cohort knowledge transfer and explore clonal relationships for enhanced immunotherapy prediction.

Main Methods:

• Proposed TMBclaw (Tumor Mutation Burden-based Clonal attention with Laplacian Adaptive Weighting), a graph-regularized multi-task learning framework.
• Integrated group-structured cohorts for unified analysis and cross-cohort knowledge transfer.
• Validated the model on four cohorts (238 patients) and external validation cohorts (1433 patients) across NSCLC, melanoma, and nasopharyngeal carcinoma.

Main Results:

• TMBclaw significantly outperformed conventional methods in prognostic accuracy and risk stratification.
• The framework systematically quantified clonal dynamics and identified driver clones.
• Demonstrated improved understanding of tumor heterogeneity and interpretable insights into immunotherapy response.

Conclusions:

• TMBclaw offers a robust approach to predict immunotherapy response by accounting for tumor clonal heterogeneity.
• The method has the potential to refine patient stratification for personalized cancer treatment strategies.
• Provides a framework for deeper insights into tumor evolution and immune response in cancer patients.