PRVCNN

Predicting regulatory variants with Convolutional Neural Network

Gaoxiang Li

Auburn University

Advisor:Li Chen, Xiao Qin

How to use:

(1)After download and unzip the file, run: python PRVCNN.py in this directory. command: python PRVCNN.py --h (for help to show all command line instruction)

1.training:

         --train       training model
         --model_name  the model name you want to save default:PRVCNN

Path for training input

         --training_feature  default='./data/training_feature_all.csv'
         --training_label    default='./data/training_label_all.csv'

2.testing:

         --test       testing process
         --test_name    the model name you want to test default:PRVCNN

Path for testing input:

         --testing_feature   default='./data/t1_imbanlance_feature.csv'
         --testing_label     default='./data/t1_imbanlance_label.csv'
         --result_score_file default='score1.csv' Result score file name you want to save

3.evaluation

         --evaluation   evaluation model
         --eval_name    the model name you want to eval default:PRVCNN
         --eval_dataset default=0 The compared dataset to be evaluated,1 for allelic imbalance'
                         '2 for eQTL, 3 for dsQTL, 4 for validated

Path for eval input

         --eval_score_file  default=default='./result/score1.csv' The score file to evaluate
         --eval_true_label  default=default='./data/t1_imbanlance_label.csv'   The true label to evaluate
         --eval_name        default='Your method'  The name of curve showed in figure

4.Parameter

          --batch_size default=16
         --max_epoch  default=2000
         --learning_rate default=0.0001
         --learning_rate_decay default=0.99
         --dropout_rate default=1
         --L2_regularizer default=0.05
         --kernel_size default=5

(2)Fpr other test: Run: python other_test.py

         --SVM   SVM method
         --RF    Random Forest method
         --training_feature    default='./data/training_feature_all.csv'
         --training_label      default='./data/ML_training_label.csv'    (only one column)
         --testing_feature     default='./data/t1_imbanlance_feature.csv'
         --testing_label       default='./data/t1_imbanlance_label.csv'
         --result_score_file   default='score1_svm.csv'   Result score file name to be saved

Prediction Score：

file score1.csv Allelic imbalance

file socre2.csv eQTL

file socre3.csv dsQTL

file score4.csv validated regulatory

Training code: PRVCNN_train

Testing code: PRVCNN_test1-4

notes:input must be reshaped and change the string feature to one-hot code

GaoxiangLi/PRVCNN

PRVCNN