Search

Additional file 1: Figure S1. Illustration of different types of Locus Definitions (LocDefs) used in this study. ChIP-seq peaks (top) are assigned to genes if they are located within a chosen LocDef, including: “nearest TSS”, “5kb to TSS”. Figure S2. Bar plots of F1 scores for each cell type and TF among the evaluation and testing ChIP-seq data sets. Each panel represents one of the top 10 EnTDefs. Cell types in the evaluation dataset are greyish, while those in the testing dataset are bluish. Figure S3. Characteristics of testing ChIPseq dataset and the performance of EnTDef on a completely different ChIPseq dataset. (A) Boxplots of the number of peaks in the evaluation and testing ChIP-seq data. (B) The correlation between the number of peaks in evaluation/testing ChIP-seq datasets (log2 scale) and the average F1 score across the top 10 best EnTDefs. (C) Scatter plot of average AUPRC (aura under the precision-recall curve) vs AUROC (aura under the receiver operating characteristic curve) of the top 10 EnTDef, top 10 EnTDef_plus5kb and baseline locus definitions (nearest_tss and 5kb_outside) in the gene set enrichment (GSE) testing on the 31 independent ChIP-seq datasets from 14 transcription factors in 9 cell lines, which were completely different from the ones used in the EnTDef evaluation analysis. Figure S4. Correlation of average F1 scores for a TF across EnTDefs. (A) The correlation between average F1-scores calculated on a TF in a particular cell type using CT-EnTDefs of the matched cell type (“same-CT.EnTDef” on x-axis) and the ones calculated on the same TF using general EnTDefs (“general.EnTDef” on y-axis). (B) The correlation between average F1-scores calculated on a TF in a particular cell type using CT-EnTDefs of a different cell type (“diff-CT.EnTDef” on x-axis) and the ones calculated on the same TF using general EnTDefs (“general.EnTDef” on y-axis). Each dot represents an average F1-score of a TF across EnTDefs, and each panel is one of four cell types (GM12878, H1HESC, K562 and MCF7) for which the CT-EnTDefs were created and evaluated respectively. Figure S5. Correlation of F1 scores for each TF and EnTDef pair. (A) The correlation between F1-scores calculated on a TF in a particular cell type using a CT-EnTDef of the matched cell type (“same-CT.EnTDef” on x-axis) and the ones calculated on the same TF using general EnTDefs (“general.EnTDef” on y-axis). (B) The correlation between F1-scores calculated on a TF in a particular cell type using a CT-EnTDef of a different cell type (“diff-CT.EnTDef” on x-axis) and the ones calculated on the same TF using a general EnTDef (“general.EnTDef” on y-axis). (C) The correlation between average F1-scores calculated on a TF in a particular cell type using a CT-EnTDef of the matched cell type (“same-CT.EnTDef” on x-axis) and the ones calculated on the same TF using a CT-EnTDef of a different cell type (“diff-CT.EnTDef” on y-axis). Each dot represents a F1-score of a TF and EnTDef pair, and each panel is one of four cell types (GM12878, H1HESC, K562 and MCF7) for which the CT-EnTDefs were created and evaluated respectively. Figure S6. Different views of sensitivity and specificity of the selected EnTDefs and baseline locus definitions as compared to the BENGI dataset. (A) Violin plots of average sensitivity and specificity of EnTDefs (top 10 EnTDefs.plus5kb [EnTDef.top_plus5kb], top10 EnTDefs, middle 10 EnTDefs and bottom 10 EnTDefs) in each type of BENGI datasets. The values of the top1 EnTDef were annotated by a red star and the values of baseline locus definitions (nearest TSS [nearest_tss] and >5kb [5kb_outside]) were annotated by the dashed lines (red: nearest TSS, blue: >5kb). ANOVA tests were performed between top10 EnTDefs and nearest TSS: sensitivity in BENGI with fixed positive/negative ratio (BENGI _fixedRatio): p value = 3.92×10−226; sensitivity in BENGI with natural positive/negative ratio (BENGI _naturalRatio): p value =8.56×10−237; specificity in BENGI _fixedRatio: p value = 2.28×10−156; specificity in BENGI _naturalRatio: p value = 4.48×10−218 (B) Scatter plots of overall false positive rate (1-specificity) vs. true positive rate (sensitivity) of each type of EnTDefs and baseline locus definitions in BENGI _fixedRatio and BENGI _naturalRatio dataset separately. Figure S7. Comparisons of the GSE performance for the top10, middle 10 (ranked at 732 to 741) and bottom EnTDefs in two case studies. (A) Distribution of average AUC values of EnTDefs in each group when recapitulating the biological process changes between IMU and KRT HPV(+) subgroups in head and neck cancer patients which were identified in the transcriptional data generated in parallel. (B) Distribution of average AUC values of EnTDefs in each group when recapitulating the biological process changes by SOX17 overexpression in hemogenic endothelium which were identified in the transcriptional data generated in parallel. Figure S8. Illustration of interceding gene definition. The interceding genes were defined as the genes with any part of the gene body falling in the query region (i.e., the genomic regions between the farthest positions of an enhancer and its target gene pair). Figure S9. Comparison between Poly-Enrich approximate and exact methods. (A) The representative GSE result of transcription factor (TF) EGR1 in the K562 cell line: all GO terms are generally concordant with the score test (PE.Approx method) being slightly more conservative, but the depleted GO terms tend to deviate more from the PE.Exact method. (B) The correlation of F1 score of the top 19 best EnTDefs between the GSE result derived from the PE.Approx method and that derived from the PE.Exact method. Each dot represents a GSE result of a particular TF using one of the 19 EnTDefs. Figure S10. Illustration of positive and negative GOBP terms assigned to TFs by GO with the number of annotated genes ≥ 15 and ≤ 2,000. (A) Positive terms include the lowest level of assigned GOBP (leaf terms) and all of their ancestors; negative terms include the terms outside of positive ones, excluding the leaf terms and their siblings and offspring of the assigned terms. (B) More conservative sets of positive and negative terms: positive terms include assigned leaf terms and their parent and grandparent terms, excluding the ones assigned by IEA (automatically assigned, inferred from Electronic Annotation); negative terms include the terms outside of the positive ones, excluding the ancestors of positives, siblings of positives’ ancestors, and offspring of positives.

Additional file 3. Peer review history.