Training one model per TF family:
python cross_validate.py
Plots will be saved to report/.
Training one model for all TF families:
python cross_validate_one_model.py
Plots will be saved to report/one_model/.
TODOs:
-
other low throughput experimental data from the paper as test data?
-
Nested CV, actual test data
-
TF DNA-binding site sequence as input, instead of multi-head output
-
filter visualization
-
concentration as input to the model, or train as different output?
-
make sure there is no duplicated sequence across TF families, if there are, merge the data instead of concatenating (duplicating training example)