Fig. 2: Evaluation of TRAPT and competing methods on TR knockdown/knockout and TF binding datasets.

a (1) The number of TFs accurately identified by different methods, where the x-axis represents the number of target TFs ranked within predictions of the top 10 by each method, and the y-axis represents the different methods considered: TRAPT, Lisa, BART, i-cisTarget, and ChEA3. (2) Line graph depicting the accurate prediction of TFs in knockdown/knockout experiments by various computational models, where the x-axis represents the number of target TFs ranked within predictions of the top N by each method. The upper-left corner shows the area under the curve (AUC) for each method. (3) Bar graph showing the MRR scores of the TFs, with higher scores reflecting superior performance. b, c Subsequent panels maintain the formats of the panels (a), and extend the analysis to TcoFs and CRs to demonstrate the predictive capability and accuracy of each method. d The MRR scores for protein families from the TR knockdown/knockout datasets, with red indicating the upregulated set and blue denoting the downregulated set. The intensity of each color signifies the magnitude of the score. e Assessment of the performance of three methods on TR target genes from the KnockTF benchmark dataset (n = 1140), by using only the TR background library derived from Cistrome. f Assessment of the performance of five methods on TR target genes from the Lisa benchmark dataset (n = 124), using only the TR background library derived from Cistrome. The box plot illustrates the scaled ranks of the target TRs according to different models. Middle line inside each box represents the median, upper and lower bounds of the box represent the third and first quartiles, respectively. P-values are calculated by the two-sided T-test without adjustments.