What is logloss-beraf?

A tool for selection of a limited number of informative DNA methylation regions (i.e. sites) based on a combination of several feature selection methods and an ensemble-based classifier. It is expected to handle higly unbalanced and heterogeneous data. Also it is intended for the design of diagnostic panels that can be potentially used in routine laboratory practice.

Pre-requisites

It is implied that one has python 2.7 and at least a miniconda distribution installed

Quick start

1. Install logloss-beraf with all the dependencies:

conda install scikit-learn matplotlib
pip install logloss-beraf

2. Make a test run. It uses data included to the package:

logloss_beraf test_run

3. Prepare input feature and annotation tables (in CSV format). The order of samples in those tables is supposed to be the same

     # Methylation data
                Feature_1 Feature_2  Feature_3
     Sample_0   0.909642  0.823715   0.069785
     Sample_1   0.564799  0.199724   0.840741
     Sample_2   0.685081  0.489773   0.286591
     Sample_3   0.810637  0.006836   0.888038
     Sample_4   0.124098  0.347752   0.954853

     # Annotation data
     Sample_Name  Type
     Sample_0     Benign
     Sample_1     Pathologic
     Sample_2     Benign
     Sample_3     Benign
     Sample_4     Pathologic

4. Train model

logloss_beraf train \
    --features path_to_feature_table \
    --features_max_num 10 \
    --min_beta_threshold 0.2 \
    --annotation path_to_annotation_table \
    --sample_name_column "Sample_Name" \
    --class_column "Type" \
    --output_folder path_to_output_folder

5. Apply trained model to independent dataset

logloss_beraf apply \
  --features path_to_test_feature_table \
  --model path_to_trained_model
  --output_folder path_to_output_folder