Fig. 1
From: Identifying piRNA targets on mRNAs in C. elegans using a deep multi-head attention network
The overview of the devised deep multi-head attention network for identifying piRNA–mRNA binding events. The network can be divided into three different parts. First, the site-by-site motif features of the given piRNA and mRNA segment are extracted by convolution and squeezing-excitation operations. Then the piRNA binding rules are mimicked by the multi-head attention operation. Finally, the piRNA binding mRNA targets are identified using the attentive feature vectors. The pair (x, y) in the graph indicates the dimension of the operation output matrix