SConES

(Selecting Connected Explanatory SNPs)

Summary

MATLAB code implementing the method described in:

C.-A. Azencott, D. Grimm, M. Sugiyama, Y. Kawahara and K. Borgwardt (2013) Efficient network-guided multi-locus association mapping with graph cuts, Bioinformatics 29 (13), i171-i179 doi:10.1093/bioinformatics/btt238

Related work and repositories

• EasyGWAS: SConES has been integrated to EasyGWAS, a framework for the analysis and meta-analysis of GWAS data. In particular, this offers a Python interface.

• sfan: Regarding the feature selection part (i.e. after the GWAS data has been processed and the SNP scored), sfan uses a different (faster) maxflow solver, is written in Python, and also incorporates the multi-task version proposed in

M. Sugiyama, C.-A. Azencott, D. Grimm, Y. Kawahara and K. Borgwardt (2014) Multi-task feature selection on multiple networks via maximum flows, SIAM ICDM, 199-207 doi:10.1137/1.9781611973440.23

• MultiSConeS: For the original version of this multi-task version, see Multi-SConES.

Demo:

In the code folder, there is a MATLAB script demo.m.

To run the demo start MATLAB and type in:

demo()

Data:

• X = Genotypematrix of size n x s, where n is the number of samples and s is the number of SNPs
• Y = Phenotypevector of size n x 1, where n is the number of samples
• W = sparse network with size s x s

Demo files are provided in the data folder.

<<<<<<< HEAD

SConES:

=======

Running SConES for given values of the lambda and eta parameters

[indicators, objectives] = scones(data, option)

Input:

a957a00a7dfd4d289aeb55fba4d65e96b145adba To run SConES two parameters are needed. The first one is a data cell array:

• data.X is the genotype data
• data.Y is the phenotype
• data.W is the sparse network
  <<<<<<< HEAD
• data.selected_PCs is the number of principle components that should be used for population structure correction
• data.lambda_values is a vector of size 1 x k with k values for lambda
• data.eta_values is a vector of size 1 x h with h values for eta

=======

• data.selected_PCs is the number of principal components that should be used for population structure correction
• data.lambda_values is a vector of size 1 x k with k values for lambda
• data.eta_values is a vector of size 1 x h with h values for eta

a957a00a7dfd4d289aeb55fba4d65e96b145adba The second parameter is a options cell array (optional - default values are specified):

• options.automatic : if this parameter is true data.lambda_values and data.eta_values are determined automatically (default: true)
• options.number_parameters : this parameters specifices the number of eta and lambda values in the case options.automatic is set to true (default: 10)
• options.stdout : if this parameter is true output is printed into the terminal window (default: true)
  <<<<<<< HEAD
• options.nfold : if scones_crossvalidation is called this parameter specifices the number of folds (default: 10)
• options.seed : if scones_crossvalidation is called this parameter specifies a seed for splitting the data (default: 0)
  =======

a957a00a7dfd4d289aeb55fba4d65e96b145adba

Output:

• indicators = indicator matrix of size n x k x h, where n is the length of vector c, k the length of vector lambda_values and h the length of vector eta_values
• objectives = matrix with all objective values with size k x h for the grid of lambda x eta values

<<<<<<< HEAD

=======

Using an inner cross-validation to determine the best values of the lambda and eta parameters

[indicators, objectives] = scones_crossvalidation(data, option)

Input:

The first parameter (data) is the same as described above.

The second parameter (options) can additionally take the following values:

• options.nfold : if scones_crossvalidation is called this parameter specifices the number of folds (default: 10)
• options.seed : if scones_crossvalidation is called this parameter specifies a seed for splitting the data (default: 0)

a957a00a7dfd4d289aeb55fba4d65e96b145adba

Contact

Any questions can be directed to Chloe-Agathe Azencott: chloe-agathe.azencott [at] mines-paristech.fr