To train a deep learning model for phenotype prediction, we take samples from fasta files and assign labels to the sequence based on a table with lable annotations.
Summary of a model architecture for spore training:
We use very similar architectures for cell size, pathogenicity and oxygen requirements training. Depending on the task, we have to adjust the loss function and activation of the output layer.
Models can also have multiple output layers, i.e. the model tries to predict several features at once. For the morphology prediction, the model predicts cell shape, flagellum, gram staining and motility.
