Skip to main content

Advertisement

Springer Nature is making Coronavirus research free. View research | View latest news | Sign up for updates

Fig. 5 | BMC Bioinformatics

Fig. 5

From: Visualizing complex feature interactions and feature sharing in genomic deep neural networks

Fig. 5

Distribution of positive sample activation level, negative sample activation level and motif matching p-values of filters grouped by their ONIV score ranking. We collected convolutional filters from all 422 TF binding models and group them into four groups by the ranks of ONIV score, each containing 1688 filters. Each panel represents one of the groups and the ONIV ranks increase from left to the right. The averaged activation scores across all negative and positive examples are calculated for each filter, and is normalized to [0,1] within each network. The top ranking group (right most) has high activation in positive examples while low activation in negative examples, and has the most significant motif matching pvals. This is suggesting that DeepResolve ranks highly relevant and informative filters that can separate positive and negative set well

Back to article page