A slightly modified version of the code from https://github.com/OrensteinLab/DeepZF/, with a few additional files.

The code requires protein-bert, and works on python 3.6.8 with keras 2.6.0 and tensorflow 2.6.2

Assemble the pretrained model model.p for BindZFpredictor by concatenating its splits in the folder BindZF_predictor/code/

cat x?? > model.p

Call BindZFpredictor on a csv file containing (a description of) the zinc fingers (and their context) and get the predictions in a tsv file with the following commands (the first one for the human ZFs, the second one for the mouse ZFs) :

python3  BindZF_predictor/code/main_bindzfpredictor_predict.py -in Zkscan3_exps/Zkscan3_ZFs_df_human_for_binding_predictions.csv -out Zkscan3_exps/Zkscan3_binding_preds_human.tsv -m BindZF_predictor/code/model.p -e BindZF_predictor/code/encoder.p -r 1

python3  BindZF_predictor/code/main_bindzfpredictor_predict.py -in Zkscan3_exps/Zkscan3_ZFs_df_mouse_for_binding_predictions.csv -out Zkscan3_exps/Zkscan3_binding_preds_mouse.tsv -m BindZF_predictor/code/model.p -e BindZF_predictor/code/encoder.p -r 1

The file turn_sequence_into_csv_for_binding_pred.py can help you create csv files in the correct format.

Predict the PWM from csv files containing descriptions of the zinc fingers in the right format (first line for the human ZFs, second one for the mouse ZFs):

python3.6 PWMpredictor/code/main_PWMprecictor.py -in Zkscan3_exps/Zkscan3_ZFs_df_human_for_PWM.csv -out Zkscan3_exps/predictions_Zkscan3_human_156_model.txt -m Zkscan3_exps/transfer_model_final.h5

python3.6 PWMpredictor/code/main_PWMprecictor.py -in Zkscan3_exps/Zkscan3_ZFs_df_mouse_for_PWM.csv -out Zkscan3_exps/predictions_Zkscan3_mouse_156_model.txt -m Zkscan3_exps/transfer_model_final.h5

The file prepare_sequence_for_PWM_prediction.py can help you create the csv files in the correct format.