Figure 1:

(A) An unrolled sketch of the NN architecture. The circles at the bottom represent the time series of raw signal input data. Local pattern information is then discriminated from this input by a CNN. The output of the CNN is then fed into an RNN to discern the long-range interaction information. A FC layer is used to get the base probability from the output of the RNN. These probabilities are then used by a CTC decoder to create the nucleotide sequence. The repeated component is omitted. (B) Final architecture of the Chiron model. Variants of this architecture were explored by varying the number of convolutional layers from 3 to 10 and recurrent layers from 3 to 5. We also explored networks with only convolutional layers or recurrent layers, 1×3 conv, 256, no bias means a convolution operation with a 1×3 filter and a 256-channeloutput with no bias added. LTSM = long-term short memory.