This code base provides all the necessary pieces to reproduce the main results of the polyAcode manuscript. If you have any questions, please email [email protected]

• Python (2.7). Python 2.7.6 is recommended.

• Numpy(>=1.6.1).

• Scipy(>=0.10).

• Scikit-learn(>=0.16.1).

• Shogun(>=3.2.0). Required for wdsvm.py

features_annotation.xls contains detailed annotation about the 658 features.

• polyAsites_specific.txt contains the 2276 tissue-specific PASs described in the manuscript.
• polyAsites_unspecific.txt contains the 3903 tissue-unspecific(constitutive) PASs described in the manuscript.
• features_specific.txt contains the 658 RNA features for the 2276 tissue-specific PASs.
• features_unspecific.txt contains the 658 RNA features for the 3903 tissue-unspecific PASs.
• seq_specific.txt contains the +/- 100nt sequence for the 2276 tissue-specific PASs.
• seq_unspecific.txt contains the +/- 100nt sequence for the 3903 tissue-unspecific PASs.
• data_features_train.npy is the numpy format for training lr.py, lsvm.py adaboost.py. It is a 4000x659 matrix with 4000 corresponding to 2000 tissue-specific PASs and 2000 tissue-unspecific PASs, and 659 corresponding 658 features and the last column representing the label(+1 for tissue-specific and -1 for tissue-unspecific).
• data_features_test.npy is the numpy format for testing lr.py lsvm.py adaboost.py. It is a 552x659 matrix with 276 tissue-specific PASs and 276 tissue-unspecific PASs.
• data_features_Bcell_train.npy, data_features_Bcell_test.npy are similarly defined as above but only for Bcell-specific PASs.
• data_features_Testis_train.npy, data_features_Testis_test.npy are similarly defined as above but only for Testis-specific PASs.
• data_seq_train.txt is the +/- 100nt sequence with the label for training wdsvm.py. It contains 2000 tissue-specific PASs and 2000 tissue-unspecific PASs.
• data_seq_test.txt is the +/- 100nt sequence with the label for testing wdsvm.py. It contains 276 tissue-specific PASs and 276 tissue-unspecific PASs.

• features_specific.txt contains the 658 RNA features for the 706 tissue-specific PASs.
• features_unspecific.txt contains the 658 RNA features for the 1114 tissue-unspecific PASs.
• data_features_train.npy is the numpy format for training lr.py, lsvm.py adaboost.py. It is a 1000x659 matrix with 1000 corresponding to 500 tissue-specific PASs and 500 tissue-unspecific PASs, and 659 corresponding 658 features and the last column representing the label(+1 for tissue-specific and -1 for tissue-unspecific).
• data_features_test.npy is the numpy format for testing lr.py lsvm.py adaboost.py. It is a 412x659 matrix with 206 tissue-specific PASs and 206 tissue-unspecific PASs.

Training AdaBoost model is done by adaboost.py. For example:

$ ./adaboost.py ./MayrLab/data_features_train.npy ./MayrLab/data_features_test.npy 0 10 5

In this case, 0 is the feature index; 10 means using 10-fold cross-validation; 5 means using 5 process for training. The complete feature index is given below:

Training logistic regression model is done by lr.py. For example:

$ ./lr.py ./MayrLab/data_features_train.npy ./MayrLab/data_features_test.npy 0 10 5

which is similarly defined as adaboost.py.

Training linear SVM model is done by lsvm.py. For example:

$ ./lsvm.py ./MayrLab/data_features_train.npy ./MayrLab/data_features_test.npy 0 10 5

which is similarly defined as adaboost.py.

Training WD-SVM model is done by wdsvm.py. For example:

$ ./wdsvm.py ./MayrLab/data_seq_train.txt ./MayrLab/data_seq_test.txt 10

In this case, 10 means using 10-fold cross-validation;