Figure 2.
The dot plot illustrates the SHAP values representing feature contributions to a machine learning model's output for genetic diagnosis in IEI. The horizontal axis represents SHAP values, indicating the impact on model output, ranging from minus 3 to 3. The vertical axis lists various features such as chest or abdominal pain, family history, autoantibody, lymphocytes counts, pre-diagnosis, skin rash, enterocolitis, serum IgG level, headache, female sex, arthralgia, age onset, autoinflammatory, and infection. Each dot represents one patient, with color indicating feature value: red for high and blue for low. Positive SHAP values suggest increased contribution to model outputs predicting a final genetic diagnosis, while negative values indicate reduced contribution. Notable patterns include family history and lower lymphocyte counts showing relatively positive contributions, whereas chest or abdominal pain tends to contribute negatively. The plot reveals the varying impacts of different features on the model's diagnostic predictions.
SHAP-based interpretation of a machine learning classifier for genetic diagnosis in IEI. SHAP summary plot illustrating feature contributions to CatBoost model output for a final genetic diagnosis consistent with IEI. Each dot represents one patient; color indicates feature value (red = high, blue = low). Positive SHAP values indicate increased contribution to model outputs predicting a final genetic diagnosis, whereas negative values indicate reduced contribution. Family history and lower lymphocyte counts showed relatively positive contributions, while chest or abdominal pain tended to contribute negatively.