Fig. 1
From: Human mitochondrial genome compression using machine learning techniques
The architecture of the DeepDNA model. Firstly, the input genome sequence is transformed into one-hot 4-dimensions bit matrix; A convolution layer activated by a rectified linear units acts as a local feature extractor, its output is a matrix with column matrix of the convolution filter and the row matrix of the position in the input sequence; A max-pooling procedure is used to reduce the size of the output matrix and only preserve the main features; The subsequent Long Short-Term Memory network (LSTM) layer is considered as acting the role of capturing sequence long-term features; A flattened fully connected layer is to collect LSTM outputs; The last layer performs a sigmoid non-linear transformation to a vector that serves as probability predictions of the sequence base