Figure 2.

OGT prediction models. Input tRNA sequences are one-hot encoded and padded with 0s to make the matrix. (A) The general structure of the temperature classifier model. The model has two channels. Both channels are fed a tRNA. Each channel starts with a convolutional layer and a maximum pooling layer. The sequence is flattened and passed to two fully connected and dropout layers. The two channels are concatenated and then passed to fully connected layers. The output layer has two neurons for binary classification. (B) The general structure of the regression model. The input layer is followed by two convolutional layers and a maximum pooling layer. Then, data are processed through fully connected dense layers, resulting in a single OGT prediction for each tRNA. This figure shows the general structure only, and the exact number of layers are selected with hyper-parameter optimization. Selected hyper-parameters are provided in the Supplementary Files.