Table 2 Performance of algorithms to predict the response of cancer cell lines to drugs.

Data type	Model	Features	ACC	PPV	TPR	MCC	F-score
Matrix	MLP	DGE, LE	0.55	0.63	0.64	0.27	0.55
Matrix	MLP	DGE, KIP, DGA	0.60	0.60	0.60	0.20	0.60
Matrix	SVM-PCA	DGE, LE	0.61	0.70	0.51	0.10	0.40
Matrix	SVM-PCA	DGE, KIP, DGA	0.62	0.72	0.53	0.16	0.45
Matrix	RF-PCA	DGE, LE	0.42	0.56	0.52	0.06	0.33
Matrix	RF-PCA	DGE, KIP, DGA	0.44	0.56	0.53	0.09	0.39
Graph	WL Tree	DGE, KIP, DGA	0.68	0.67	0.65	0.32	0.65

A graph-based approach is compared to two matrix-based methods. The performance of each algorithm is cross-validated at the tissue level.
MLP multilayer perceptron, SVM-PCA Support Vector Machines with Principal Component Analysis, RF-PCA Random Forest with Principal Component Analysis, WL Tree Weisfeiler–Lehman graph kernel, DGE differential gene expression, LE ligand embeddings, KIP kinase inhibitor profiling, DGA disease-gene associations, ACC accuracy, PPV precision, TPR recall, MCC Matthews correlation coefficient.

Quick links

Search