Tensorflow 2.1+ compatibility version of deepC. Adapted from the 1.8.0 version. Runs in Tensoflow2 (2.1+ with CUDA 10.1). The code is only adapted to load in tf1 compatibility mode. We are working on a reimplementation in tensorflow2 style.
-
deepCregr.py model implementation of deepC. Flexible number of convolutional and dilated convolutional layers. Residuals (in the dilated layers) and batch normalization can be turned on.
-
deepCregr_utility.py model implementation with more flexible intermediate outputs mainly for saliency computation
-
run_training_deepCregr.py script for training a deepC model, requires formated/pre-rpocssed data such as the provided ones and a link to the matching reference genome.fa and .fai file
-
run_deploy_shape_deepCregr.py script to run prediction from sequence. Requires a trained deepC model and a bed like file with chrom start end replace in a tab separated file with bed 0-based coordinate encoding, replacer being the sequence you want to exchange the respective genomic window for. Use reference if you want to run on the reference sequence.
-
run_deploy_shape_combination_deepCregr.py same as above but applies all variants listed in the input file to the sequence before running the prediction. For multiple variants.
-
run_get_saliency.py script for calculating the saliency with respect to input.
If help messages for command line arguments don't display, please have a look at the script.