The targets identified at P = 0.95 should be the most relevant hits. Regarding Dr.

The targets identified at P = 0.95 should be the most relevant hits. Regarding Dr. Ideker’s comments on comparisons to other methods, we feel that showing statistical significance and biological interpretations is sufficient for a publication on this topic. As this method continues to develop, a more detailed comparison with other methods is desirable, and we hope to complete that in future work. We believe that the significance of this work lies not just in method, but in a combination of methodology and biology; for the latter, especially the insights on the important regulator, Wt1. Finally, regarding the writing and organization of the manuscript, we realize that there is content in the introduction which is methodological in nature and that there is also some order Tasigna background details given in the results. Given the nature of this topic and the fact that we are attempting to appeal to both computationally oriented and biologically oriented audiences, we felt the need to repeat certain methodological points which are important. We further felt that the biological meaning of the data presented in the Discussion section has greater depth given the detailed discussion of WT1. Moving this section to the Introduction may break the continuity of the story.Reviewer’s report 2 Vladimir A. Kuznetsov, Division of Genome and Gene Expression Analysis Bioinformatics Institute, Singaporemotif datasets and some other sequence information mapped on human genome as training set of SVM algorithm to predict new targets for selected TFs. Based on their counting of occurrence of several types of DNA fragments (e.g. motifs, preferred patterns of bases, evolutionarily conserved DNA fragments etc.) in RefSeq genes or their putative promoter regions the authors predicted 933 targets for 152 TFs, including 354 target genes for WT1 TF. An association of predicted OCT4 gene targets with Wnt pathway and some other biological and clinical correlations were considered. Main Comments Evaluation and control of accuracy, specificity, sensitivity of gene target predictions and consistency of the predictions with previous studies are my major focuses in consideration of the work. I concern regarding the predictive power of TFSVM methodology, performance of the method, biological significance of predicted targets, interpretation and extrapolation of the results, and independent validation. 1. At the beginning of the paragraph “Results and Discussion” the authors claim that cross-validation performance measures of their method are determined at the decision threshold of P = 0.5 and they further explain the reasons of their choice. That means that only genes with average scores higher than this threshold have a chance of being true targets. PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/27362935 However, this P = 0.5 threshold might still be a small or a large number depending on the situation. Since P is an average score it is also sensitive to outliers. It may be better preferable to base selection on a criterion that also takes into account the variance of the 100 P scores for each gene or alternatively to use the additional information of how many times (percentage) is P exceeding a threshold over the 100 iterations.[Authors’ Response] Our use of an average P = 0.5 threshold is due to the very observation that there may be variability between measurements. Since the choice of a negative set is random, and therefore potentially noisy, we felt that forcing an average cutoff is clearly better than any single classifier measurement.