The Next Generation of Transcription Factor Binding Site Prediction
Figure 4
Performances comparison between 0-order TFFMs, other TFFMs, and weight matrices.
For the 96 ChIP-seq data sets used in Figure 3 (using genomic background), the ratio between the AUC value using a specific model and the best AUC obtained is plotted. (A) The three types of TFFMs were used (1st-order, detailed, and 0-order TFFMs). AUC ratios are ranked from the least to the most favourable to the 1st-order and detailed TFFMs. We observe that the 1st-order and detailed TFFMs outperform the 0-order TFFMs when discriminating ChIP-seq sequences from genomic background sequences. (B) 0-order TFFMs and WMs were used. AUC ratios are ranked from the least to the most favourable to the 0-order TFFM. We observe that the WMs outperform the 0-order TFFMs when discriminating ChIP-seq sequences from genomic background sequences.