Comparison to other methods.

Most of tFIRE prediction is better than others. Except the use of AlignACE, BioProspector for GAL4 as well as the use of MEME for HAP1 outperform our structure-based approach.aNS: Number of sites collected from TRANSFAC and SCPD.bNO: Number of ORFs these binding sites taking place.cSE: Sensitivity [TP/(TP+FN)]. SP: Specificity [TP/(TP+FP)]. AUC: Area Under Receiver operating characteristic Curve.dAlignACE [48] v4.0 result using parameter as: number of columns = LTFBS.eBioProspector [49] result as: motif width = LTFBS; top motifs to report = NSites.fMEME [47] v4.3.0 result as: maximum motif width = LTFBS; maximum sites = NSites; maximum motif number = NSites.gAverage value over LEU3, GAL4, MCM1, MATALPHA2, TBP, PUT3, GCN4, HAP1, RAP1.多数实验场景下，tFIRE的预测性能优于其他同类方法；仅针对GAL4的AlignACE、BioProspector分析，以及针对HAP1的MEME分析，其性能优于本研究的基于结构的方法。 aNS：从TRANSFAC与SCPD数据库中收集的结合位点总数。 bNO：这些结合位点所在的开放阅读框（Open Reading Frame，ORF）数量。 cSE：灵敏度（Sensitivity），计算公式为TP/(TP+FN)；SP：特异度（Specificity），计算公式为TP/(TP+FP)；AUC：受试者工作特征曲线下面积（Area Under Receiver Operating Characteristic Curve）。 d：AlignACE [48] v4.0的运行结果，所用参数为：列数=LTFBS。 e：BioProspector [49] 的运行参数为：基序（motif）宽度=LTFBS；待报告的排名靠前的基序数量=NSites。 f：MEME [47] v4.3.0的运行参数为：最大基序宽度=LTFBS；最大结合位点数=NSites；最大基序数量=NSites。 g：针对LEU3、GAL4、MCM1、MATALPHA2、TBP、PUT3、GCN4、HAP1、RAP1的平均计算结果。