enkelejdamiho / deepid Goto Github PK

A deep learning model for accurate diagnosis of infection using antibody repertoires

1. Running DeepID requires the python (3.7 version or later) runtime environment;
2. Make sure that extension package including Numpy, Pandas and paddlepaddle 1.8.4 have installed for current python environment;
3. Download the RLM.pdparams, SLM.pdparams, RLM.py and SLM.py, DeepID.py, test_repertoire_level_features.npy, test_sequence_level_features.npy and y_test.npy to the running directory;
4. The command for evaluating RLM on the test_repertoire_level_features is:

   python RLM.py input_x_file input_true_Y output_file
   

   For example:

   python RLM.py test_repertoire_level_features y_test RLM_test_rlt.csv
   

5. The command for evaluating SLM on the test_sequence_level_features is:

   python SLM.py input_x_file input_true_Y output_file
   

   For example:

   python SLM.py test_sequence_level_features y_test SLM_rlt_4feat.csv
   

6. The command for evaluating DeepID on the test set is (Must run RLM and SLM first):

   python DeepID.py input_RLM_rlt input_SLM_rlt output_file
   

   For example:

   python DeepID.py test_rlt.csv rlt_4feat.csv DeepID_rlt.csv
   

7. The command for muti-classification on Healthy, HBV, influenza, and COVID-19:

   python muti-classification.py test_muti-classification y_test_muti-classification output_file
   

1. The RLM.pdparams and SLM.pdparams are the model files that have been trained;
2. RLM.py, SLM.py and DeepID.py are the scripts for RLM, SLM and DeepID model;
3. The test_repertoire_level_features.npy and test_sequence_level_features.npy are the 547 repertoire-level features and 160 sequence-level features for test dataset, respectively. The names and order of these features are listed in the Feature names.xlsx. test_repertoire_level_features.npy is a 120*547 matrix with 120 samples and 547 features; test_sequence_level_features.npy is a 120*160*160 matrix, in which the three dimensions are samples, clones and sequence_level_features, respectively;
4. The y_test.npy is the true labels of the test dataset and is only used for accuracy calculation;
5. RLM_test_rlt.csv, SLM_rlt_4feat.csv and DeepID_rlt.csv are output files of the RLM.py, SLM.py and DeepID.py, respectively. The five columns of the CSV files are Probability of 0 (infection), Probability of 1 (healthy), samples are predicted to be class 0, samples are predicted to be class 1 and the true labels.
6. The user also can apply the DeepID to their test dataset by replacing the input data (test_repertoire_level_features.npy, test_sequence_level_features.npy and y_test.npy). 7）For the muti-classification, 0：COVID-19; 1:influenza; 2->HBV; 3:Healthy.

Chen Y, Ye Z, Zhang Y, et al. A Deep Learning Model for Accurate Diagnosis of Infection Using Antibody Repertoires[J]. The Journal of Immunology,20222,08 (12): 2675–2685.

If you have any questions or problems, please e-mail [email protected]