SigMod: an exact and efficient method to identify a strongly interconnected disease-associated module in gene network
Motivation: Apart from single marker-based tests classically used in genome-wide association studies (GWAS), network-assisted analysis has become a promising approach to identify the joint effect of multiple genetic factors on disease. To date, most network-assisted methods aim at finding genetic units connected in a background network, whatever their density or strength of connections. This can hamper the findings as sparse connections are non-robust against noise from either the GWAS results or the network resource.
Results: We developed SigMod, a novel method that integrates GWAS results with gene network to identify a gene module enriched in high association signals and tend to have strong interconnected. SigMod identifies a gene module by solving the optimization problem:
maxmize: obj(u) = c'u + \lambda u'Au + \eta |u|,
u is a vector of binary variables to be optimized.
u=1 indicates a gene is selected while
c is a vector of gene scores representing the confidence/scale of gene-disease association, which can be computed from the GWAS results using gene-based methods.
A is the adjacency matrix of the gene network.
\eta are tuning parameters. The meaning of each term in
c'u: the joint effect of all selected genes
u'Au: the overall connection strength among selected genes
|u|: the number of selected genes
Therefore the first term
c'u encourages high score genes to be selected; the second term
\lambda u'Au encourages the strong interconnection among selected genes; the third term
\eta |u| limits the number of genes to be selected. We showed that the maximization of
obj(u) can be solved exactly and efficiently through min-cut algorithms. Compared to existing methods, SigMod has several desirable properties: (i) edge weights quantifying evidence of connections between genes are taken into account; (ii) the selection path corresponds to variation of sparsity parameter can be computed rapidly; (iii) the identified gene module is strongly interconnected, hence includes genes of highly functional relevance; and (iv) it is robust against noise from either the GWAS results or the network resource. We applied SigMod to both simulated and real data. It outperforms state-of-the-art network-assisted methods in identifying disease-association genes.
The source code and user manual are included in the SigMod_v2.zip file. If you are intereseted in the method or meet any problem in your application, feel free to contact: firstname.lastname@example.org
- Author: Yuanlong LIU
- Affiliation: INSERM (French National Institute of Health and Medical Research), Genetic Variation and Human Diseases Unit, UMR-946, Paris, France
Please cite the following article if you use this method
- Yuanlong Liu, Myriam Brossard, Damian Roqueiro, Patricia Margaritte-Jeannin, Chloé Sarnowski, Emmanuelle Bouzigon, Florence Demenais; SigMod: an exact and efficient method to identify a strongly interconnected disease-associated module in a gene network. Bioinformatics 2017 btx004. doi: 10.1093/bioinformatics/btx004