Fig. 4

Model interpretation results by N-adjusted kernel transformation. a Visualization for effects of N in model0 during initiation. Values annotated in the heatmap indicate the order of N sorted by effects of 5 characters in sequences. 4 means N has the largest effect and 0 means effects of N is minimum among 5 characters. b Distribution for 4 transcription activators detected from kernel transformation on expressed (red line) and low expressed genes (blue line). c Distribution for 6 DOF genes detected from kernel transformation on expressed (red line) and low expressed genes (blue line). d Comparison among features extracted by CNN interpretation and meme. Features extracted by two methods were input into SVM model to predict transcription status. SVM models were evaluated in accuracy, recall rate, precise rate and f1 score. Bars with shadows represent for SVM build on features extracted by MEME and bars without features represent for SVM build on features detected from CNN