2′-O-methylations (2′-O-Me) are one of the most important layers of regulatory control over gene expression.
By utilizing Deep Learning, DeepOMe has developed a new algorithm for prediction of 2′-O-Me sites.
DeepOMe can predict the exact locations of 2′-O-Me by a many-to-many mode.
A many-to-many procedure is introduced to train and test.
A hybrid CNN and BLSTM architecture is utilized to automatedly extract features from primary mRNA sequences.
DeepOMe is constructed by a network with ten dilated 1D convolutional neural networks and two bidirectional LSTM layers.
Softmax function is implemented in the output fully connected layer.