Fig. 5: SHAP-based assessment of feature contributions to LUNAR recurrence predictions.
From: A deep learning model to predict glioma recurrence using integrated genomic and clinical data
The 20 most influential features according to DeepExplainer. Positive SHAP values (dots to the right) indicate an increase in the model’s prediction (towards an early prediction), while negative values (dots to the left) indicate a decrease (towards a late prediction). Note that for categorical features, red = Yes and blue = No, and the data points shown are post-processing and transformation. See the Methods and Fig. 2 for further details on the label Astrocytoma wildtype.
