Skip to main content
. 2017 Nov 16;34(8):1261–1269. doi: 10.1093/bioinformatics/btx727

Fig. 1.

Fig. 1.

Simplified sketch of a model architecture using spline transformation. In addition to DNA sequence, relative distances to various genomic landmarks are used as features. Spline transformation [Equation (1)] learns a smooth transformation of the raw distances. Transformed distances are then merged with sequence-based activations of convolutional layers