Skip to content

Latest commit

 

History

History
104 lines (73 loc) · 5.13 KB

readme.md

File metadata and controls

104 lines (73 loc) · 5.13 KB
Raw

Protein Function Prediction

  • Protein Function Prediction aims to construct models for protein function prediction. It can concate protein representations in to prepare datasets for training, testing models and making predictions.
  • We construct a model consisting of 4 steps that can be used independently or contiguously.
  • We compare it with other methods from the literature.

1. Fuse_representations:

  • This step make concatenation of protein representation vectors.

2. Prepare_datasets:

  • Concation of positive_sample_dataset and negative_sample_dataset for preparation of model dataset ( "Entry" and multi-columns representation vector )
  • Save pickle format of dataset

3. Model_training and Test:

  • Training and test for prepared data. Using models are Fully Connected Neural Network, RandomForestClassifier, SVC, KNeighborsClassifier

4. Model_prediction:

  • Make prediction for protein function You can access the information about the fuse_representations, prepare_datasets, model_training_test, prediction sub-modules of the function prediction module via the .py extension file of the relevant sub-module.

How to run Protein Function Prediction

Step by step operation:

  1. Clone repository
  2. Download datasets,unzip and place the folder
  3. Edit the configuration file : Hoper.yaml
  • i.e., python HOPER_main.py

Dependencies

You can access dependencies from case_study_env.yml file

Example of protein function prediction configuration file

parameters:
choice_of_task_name:  [prepare_datasets,model_training_test,prediction]
fuse_representations: ** # This step make concatenation of protein representation vectors.**
    representation_files: [../multi_modal_rep_ae_multi_col_256.csv,/media/DATA2/sinem/node2vec_d_50_p_0.5_q_0.25_multi_col.csv]
    min_fold_number:  2  # Minimum_number_of_combinations. For example if 3 representations are supplied and min_fold_number = 2
    #then the function will produce double and triple combinations of the protein representation vector. Such as Vec1_Vec2, Vec1_Vec3, Vec2_Vec3 and Vec1_Vec2_Vec3
    representation_names:  [modal_rep_ae,node2vec,bertavg]     
    
prepare_datasets:  
    positive_sample_data:  [../positive.csv]
    negative_sample_data:  [../neg_data.csv]
    prepared_representation_file:  [../multi_modal_rep_ae_multi_col_256.csv] 
    representation_names:  [modal_rep_ae] 

model_training_test:
    representation_names:  [modal_rep_ae]
    scoring_function:  ["f_max"]  # "f1_micro","f1_macro", "f1_weighted", "f_max"
    prepared_path:  ["../results/modal_rep_ae_binary_data.pickle"]
    classifier_name:  ["Fully_Connected_Neural_Network"] #"RandomForestClassifier", "SVC", "KNeighborsClassifier", "Fully_Connected_Neural_ Network",
prediction:
    representation_names:  [modal_rep_ae] 
    prepared_path:  ["../rep_file/rep_dif_ae.csv"]
    classifier_name:  ['Fully_Connected_Neural_Network']         
    model_directory:  ["../case_study_results/test/modal_rep_ae_Fully_Connected_Neural_Network_binary_classifier.pt"]

Definition of output files (results)

"representation_names_fused_representations_dataframe_multi_col.csv": Prepared dataset

  • Training result files:

    • "case_study_results/training/representation_names_model_name_training.tsv": training results which contains 29 columns
Column names Column names Column names
"representation_name" "classifier_name" "accuracy"
"std_accuracy" "f1_micro" "std_f1_micro"
"f1_macro" "std_f1_macro" "f_max"
"std_f_max" "f1_weighted" "std_f1_weighted"
"precision_micro" "std_precision_micro" "precision_macro"
"std_precision_macro" "precision_weighted" "std_precision_weighted"
"recall_micro" "std_recall_micro" "recall_macro"
"std_recall_macro" "recall_weighted" "std_recall_weighted"
"hamming distance" "std_hamming distance" "auc"
"std_auc" "matthews correlation coefficient" 
- "case_study_results/training/representation_model_name_binary_classifier.pt" : saved model

- "case_study_results/training/representation_model_name_means.tsv" : mean of 5 fold results

- "case_study_results/training/model_name_representation_name_binary_classifier_best_parameter.csv"

- "case_study_results/training/representation_name_model_name_binary_classifier_training_predictions.tsv"

  • Test result files:

    • "case_study_results/test/representation_names_model_name_test.tsv":contains same columns as training results

    • "case_study_results/test/representation_model_name_binary_classifier.pt" : saved model

    • "case_study_results/test/representation_name_model_name_test_means.tsv" : mean of 5 fold results

    • "case_study_results/test/model_name_representation_name_binary_classifier_best_parameter.csv"

    • "case_study_results/test/representation_name_model_name_binary_classifier_test_predictions.tsv"

  • Prediction result files:

    • "case_study_results/prediction/Representation_name_prediction_binary_classifier_classifier_name.tsv"