Fig. 3. Application of local interpretable model-agnostic explanations (LIME) for two representative cases.

A For case 1. This example is the case of a relatively higher risk for aGVHD (positive probability of 0.64), including an HLA class I antigen mismatch in the GVH direction. B For case 2. This example is the case of relatively lower risk for aGVHD (positive probability of 0.17), including no allelic mismatches in HLA-A, -B, and -DR.