Fig. 2

The overall structure of the Basecaller: A The main components of Lokatt, from the bottom up, are normalized input, the neural network, the EDHMM layer, and output. The neural network is expanded into main component layer blocks on the right side: two residual blocks, two bi-directional LSTM clocks, and a dense layer. B The inner structure of the residual block consists of three layers of 1D convolution, followed by layer-normalization, and the Swish activation, of which the outputs are taken and added with the inputs of the residual block followed by cross-layer normalization. C The inner structure of the bi-directional LSTM layer consists of two independent LSTM layers, one in the forward direction and the other in the backward direction. The outputs of the two LSTM layers are then concatenated along the feature dimension, making the output sequence the same length as the inputs