Machine Learning Analysis of Immune Cells for Diagnosis and Prognosis of Cutaneous Melanoma

Figure 4

Prognostic feature selection and risk score model. (a) The coefficient profiles of the 24 immune cells in the LASSO algorithm. (b) Random forests feature selection (RF-FS) algorithm. The lowest point of the curve represents the lowest out-of-bag (OOB) error, which indicates the best immune cells-combined signature discovered by RF-FS. (c) Support vector machine-recursive feature elimination (SVM-RFE) algorithm. The highlighted point represents the lowest error rate, which indicates the best immune cells-combined signature identified by SVM-RFE. (d) The Venn plot of immune cells in RF-FS, SVM-RFE, and LASSO methods. E-G: the risk model in the training dataset: the risk scores distributions (e); overall survival time and vital status (f); the expression value of immune cells (g). (h–j)The risk model in the testing dataset: the risk scores distributions (h); overall survival time and vital status (i); the expression value of immune cells (j). (k–m) The risk model in the GSE54467 dataset: the risk scores distributions (k); overall survival time and vital status s (l); the expression value of immune cells (m).

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)

(m)