CLIPNET fine tuned to K562

Model weights for CLIPNET models fine tuned from LCLs to K562.

These models are compatible with the scripts in the original CLIPNET repo (use the --model_dir flag). Scripts used to preprocess data and to do model training/evaluation are available at https://github.com/Danko-Lab/clipnet_k562/

Chromosomes in fold 0 were held out for all models. The 9 remaining folds were used to train models such that data fold i was withheld from training for model i (for assignments, see data_fold_assignments.csv). The models assume a 2-hot encoding of DNA (A=[2, 0, 0, 0], C=[0, 2, 0, 0], ..., Y=[0, 1, 0, 1], ...).