RC-CNN and WSCNN models are both based on this work SIL-CNN, and implemented on the Caffe ploatform.
In the RC-CNN/ directory, the script rc-cnn.py, which incorporates the reverse-complement mode, is for transforming DNA sequences to the data format that Caffe can take, whose size is N * C * H * W (N denotes the number of sequences, and C (=4) denotes the four channels {A, C, G, T}, and H (=1) denotes the height of sequences, and W denotes the length of sequences).
- Usage example:
python rccnn.py example/train.tsv example/train_target.tsv example/data/train.h5 -k 24 python rccnn.py example/test.tsv example/test_target.tsv example/data/test.h5 -k 24 - Type the following for details on other optional arguments:
python rccnn.py -h
The trainval.prototxt is for training RC-CNN, and the deploy.prototxt is for testing RC-CNN.
In the WSCNN/ directory, the script wscnn.py, which also incorporates the reverse-complement mode, is for transforming DNA sequences to the weakly-supervised data, whose size is N * C * H * W (N denotes the number of bags (sequences), and C (=4) denotes the four channels {A, C, G, T}, and H denotes the number of instances per bag, and W denotes the length of instnces).
- Usage example:
python wscnn.py example/train.tsv example/train_target.tsv example/data/train.h5 -c 79 -s 10 -k 24 python wscnn.py example/test.tsv example/test_target.tsv example/data/test.h5 -c 79 -s 10 -k 24 - Type the following for details on other optional arguments:
python wscnn.py -h
The four files (Max, Average, Linear-Regression, and Top-Bottom) correspond to the four fusion methods in this paper, where the trainval.prototxt is for training WSCNN, and the deploy.prototxt is for testing WSCNN. In the WSCNN/Top-Bottom/caffe-scripts/ directory, the scripts are the implementation of Top-Bottom Instances method on the Caffe platform.