Performance of algorithms (area under ROC curve, AUC) in the rediscovery experiment using only NEM316 genome.

This analysis evaluated the relative performance of each algorithm to rediscover virulence genes by applying stratified n-fold cross-validations with of the entire set of S. agalactiae NEM316 genes serving as test-set in each fold. Each fold of training set comprised positive and negative examples.n: number of virulence genes in the category. Singleton virulence gene categories were excluded from this analysis, as it is not possible to perform cross-validations on training sets with n = 1. All but one (labeled*) AUCs reached the statistical significance level at α = 0.05 (two-tailed Mann-Whitley U-test). At least 3 out of 4 algorithms were still significant after adjustment for multiple testing (across the family of 4 algorithms) by the Bonferroni method. Abbreviations: ADTree: alternating decision tree; IBk: nearest neighbor classifier; SVM: support vector machine; RBF: SVM with radial basis function; Poly: SVM with polynomial kernel. Refer to the methods section for the parameters used to train the machine learning algorithms. The numbers in bold face indicate the best performing algorithm for a given category.本分析评估了各算法通过分层n折交叉验证重新发掘毒力基因的相对性能，其中每一折均以无乳链球菌（S. agalactiae）NEM316的全部基因集合作为测试集。每一折的训练集均包含正样本与负样本。n：该类别中毒力基因的数量。仅含单个毒力基因的类别被排除在本次分析之外，因无法对n=1的训练集执行交叉验证。除标记为*的1项指标外，所有受试者工作特征曲线下面积（Area Under Curve, AUC）均达到α=0.05的统计学显著性水平（双侧曼-惠特尼U检验）。经邦费罗尼（Bonferroni）法对4种算法家族进行多重检验校正后，仍至少有3种算法保持统计学显著性。缩写说明：ADTree：交替决策树（alternating decision tree）；IBk：近邻分类器（nearest neighbor classifier）；SVM：支持向量机（support vector machine）；RBF：带径向基核函数的支持向量机（SVM with radial basis function）；Poly：带多项式核函数的支持向量机（SVM with polynomial kernel）。机器学习算法的训练参数详见方法部分。粗体数字代表对应类别中性能最优的算法。