Recurrent Neural Network-based Prediction of the Location of O-GlcNAcylation Sites in Mammalian Proteins

These are the datasets and model files associated with the publication Recurrent Neural Network-based Prediction of O-GlcNAcylation Sites in Mammalian Proteins. This work uses primarily RNN models to predict the presence of O-GlcNAcylation sites. The models were trained on multiple sources of literature data on protein O-GlcNAcylation based on human-selected descriptors (v1 of the datasets) or protein sequences (v3 and v5 of the datasets).

Reproducing the models and plots

The models can be recreated by downloading the datasets and running the ANN_train.py file with the appropriate flags (run python ANN_train.py --help for details).

The plots can be recreated by running the make_plot.py file with the appropriate data version as an input (python make_plot.py v1 for Fig. 1, python make_plot.py v5 for Fig. 2, python make_plot.py shap_heatmap for Fig. 3, python make_plot.py v3 for Fig. S1, python make_plot.py v5_nested for Fig. S2, and python make_plot.py v5_shap for Fig. S3).

Using the models to predict O-GlcNAcylation sites

The Conda environment defining the specific packages and version numbers used in this work is available as ANN_environment.yaml. To use our trained model, run the Predict.py file as python Predict.py <sequence> -t <threshold> -bs <batch_size>.

Alternatively, create an (N+1)x1 .csv with the first row as a header (such as "Sequences") and all other N rows as the actual amino acid sequences, then run the Predict.py file as python ANN_predict.py <path/to/file.csv> -t <threshold> -bs <batch_size>. Results will be saved as a new .csv file.

To run Shapley value predictions in addition to the model predictions, run the Predict.py file with also the -shap flag. Whatever other flags should be included as needed.