Associations_Inference

This repository proposes a new method to infer relevant associations within a set of features. It infers relevant predictors for several regression problems combining the Boruta method [1] and the Shapley values for feature importance [2]. Please refer to BorutaShap.pdf for a brief description of this method.

This repository also proposes a python implementation for the TIGRESS method which solves the same kind of problems.

Experiments

In order to illustrate our method, we provide a jupyter notebook which applies it to the DREAM4 In silico size 100 multifactorial subchallenge. We compare the results with state-of-the-art methods such as TIGRESS [3] and GENIE3 [4].

Comparison of Boruta SHap, GENIE3 and TIGRESS for the DREAM4 challenge

Data

The training data sets and the gold standard data sets for the DREAM4 challenge were downloaded here. You can find them in the folder Size 100 multifactorial.

Requirements

To run this algorithm as well as the jupyter notebook, one will need the following python packages:

boruta
joblib
matplotlib.pyplot
networkx
numpy
pandas
scikit-learn
shap

Examples

Boruta Shap method

import pandas as pd
from sklearn.ensemble import RandomForestRegressor
from algorithms.BorutaShap import Shap , BoShapNet

df = pd.read_csv('Size 100 multifactorial\DREAM4 training data\insilico_size100_1_multifactorial.tsv' , sep = '\t')
regressor = Shap(RandomForestRegressor())

BoShap = BoShapNet(regressor = regressor , responses = list(df.columns) , predictors = list(df.columns) , n_jobs = -1)
Boshap.fit(df)

print(Boshap.selections_)

TIGRESS method

import pandas as pd
from algorithms.TIGRESS import TIGRESS

df = pd.read_csv('Size 100 multifactorial\DREAM4 training data\insilico_size100_1_multifactorial.tsv' , sep = '\t')

T = TIGRESS(responses = list(df.columns) , predictors = list(df.columns))
T.fit(df , normalize = True)

print(T.scores_)

Acknowledgements

This package was created as a part of Master internship by Nicolas Captier in the Computational Systems Biology of Cancer group of Institut Curie.

References

[1] "Feature selection with boruta package" - Kursa and Rudnicki 2010
[2] "A unified approach to interpreting model predictions" - Lundberg et al. 2017
[3] "TIGRESS: Trustful Inference of Gene REgulation using Stability Selection" - Haury et al. 2012
[4] "Inferring regulatory networks from expression data using tree-based methods" - Vân Anh Huynh-Thu et al. 2010

ncaptier/Associations_Inference