Fig. 8
From: Unravelling TPX2-centered co-expression networks as key drivers of aggressive prostate cancer
Explainable machine learning. Distribution of bootstrapped SHAP values of the 22 shared DEGs between (a) M3 (prim/norm), (b) M2 (mHSPC/prim) and (c) M1 (mCRPC/mHSPC) obtained from binary machine learning classifiers. The distribution is color-coded by VST gene expression values (from blue, low expression, to red, high expression). The further out a dot is from the separating vertical line, the higher the average SHAP value across bootstrapping iterations, as indicated in the X axis, and hence the higher the impact on the model output. (d–f) Second-order SHAP values representing feature interaction predictive strength among each of the 22 shared DEGs between M3 (prim/norm), M2 (mHSPC/prim) and M1 (mCRPC/mHSPC) obtained from binary machine learning classifiers.
