Table 1 Qualitative comparison of MuTATE and benchmark modeling approaches across key features relevant to multi-endpoint disease subtyping
Feature/Model characteristic | MuTATE | CART | GLMM | Random Forests | Deep Learning | Hierarchical Multi-label Classification (HMC) |
|---|---|---|---|---|---|---|
Handles multiple endpoints of different variable types | Yes | Limited | Limited | No | No | Partial |
Interpretable decision rules | Yes | Yes | Partial | Limited | No | Limited |
Supports heterogeneous data (e.g., categorical, continuous, molecular) | Yes | Partial | Partial | Yes | Yes | Limited |
Guards against overfitting | Yes | Partial | Yes | Yes | Partial | Limited |
Scalable to high-dimensional data | Yes | Partial | Limited | Yes | Yes | Limited |
Cross-endpoint feature ranking | Yes | No | No | No | No | Partial |
Statistical significance-aware splits | Yes | No | Yes | No | No | Limited |
Clinically interpretable subtype definitions and inclusion criteria (whole model) | Yes | Yes | Partial | Limited | No | Limited |
Unified multi-endpoint model (single tree) | Yes | No | No | No | No | No |
