Predicting cancer type from tumour DNA signatures

Table 3 Overall accuracy of the SVM classifier trained using the genes proposed by Martinez et al. and the genes selected via SVM-RFE and stability selection in this study

Classification task	25-gene panel in Martinez et al.	Top 25 SVM-RFE-based SPM genes	Top 25 SVM-RFE-based SPM and CNA genes
28 cancer types of this study	30.4 %	39.0 %	67.7 %
10 cancer types of Martinez et al.	54.6 %	57.4 %	85.4 %

The classifier was tested on 1661 unseen tumour samples
CNA copy number altered, SPM somatic point-mutated, SVM support vector machine, SVM-RFE SVM recursive feature eliminatio