deepml/e196d204a497 repository overview

From DANN CADD https://cbcl.ics.uci.edu/public_data/DANN/readme⁠

⁠deepnet

deepnet uses Google protocol buffer files to set up the model architecture and establish file links. You have to change the file paths in all pbtxt files in the deepnet folder. Hopefully it is obvious enough based on the file names which ones to change.

In the following command line, trainer.py is in the deepnet package. You will need to find the file path to it. Training can take about 24 hours. We included the BEST trained models in the deepnet folder in case you want to skip the training.

python /path/to/trainer.py model.pbtxt train.pbtxt eval.pbtxt

For calculating the prediction scores which you can use for AUC analyses, run the following command lines. This will get you numpy files containing the scores for the validation, testing, and ClinVar_ESP (change the data_proto line in train.pbtxt to the cadd_rep_dq2.pbtxt file sets. You will need the file path to the extract_neural_net_representation.py script from the deepnet package.

python /path/to/extract_neural_net_representation.py cadd_3layer_tanh_BEST train.pbtxt outputdir hidden1 hidden2 hidden3 output_layer