Large-Scale Genetic Network Learning

@inproceedings{naegele2008genetic, abstract = {Learning Bayesian networks from microarray data has received increasing attention. However, most gene-expression experiments contain the expression value of thousands of genes, while most algorithms do not scale well in these large-scale domains. Thus, the user has usually to decide which genes should be included in a Bayesian network analysis, and which are excluded from it. In contrast, we propose a method that scales well to tens of thousands of genes, allowing the inclusion of all genes for learning the network. Our new algorithm, called S-DAG, splits the large set of genes into small subsets, and learns a Bayesian network for each subset independently. Afterwards, the subnetworks are combined to one large network, considering the constraint that the structure of a Bayesian network must not contain cycles. Based on benchmark data, we show that our algorithm reconstructs large-scale networks with a higher quality than the very competitive MMHC algorithm. Additionally, we apply our algorithm on a <i>S. cerevisiae</i> and a <i>homo sapiens</i> data set. The latter data set contains over 50,000 genes, which is the largest unrestrictred Bayesian network reconstructed from data, so far. Results from both data sets show that the learned networks have a high biological relevance.}, author = {N\"{a}gele, Andreas and Arndt, Holger and Dejori, Math\"{a}us }, booktitle = {ECAI'08 Workshop on Bioinformatics, Genomics \& Proteomics, an Artificial Intelligence Approach}, citeulike-article-id = {3036913}, editor = {Tsakalides, A. and Skarlas, L. }, keywords = {gene\_expression, genetic\_network, own\_publication, structure\_learning}, location = {Patras, Greece}, month = {July}, posted-at = {2008-07-23 14:56:47}, priority = {2}, title = {Large-Scale Genetic Network Learning}, year = {2008} }

TY - CONF ID - naegele2008genetic L3 - citeulike-article-id:3036913 TI - Large-Scale Genetic Network Learning T2 - ECAI'08 Workshop on Bioinformatics, Genomics & Proteomics, an Artificial Intelligence Approach CY - Patras, Greece N2 - Learning Bayesian networks from microarray data has received increasing attention. However, most gene-expression experiments contain the expression value of thousands of genes, while most algorithms do not scale well in these large-scale domains. Thus, the user has usually to decide which genes should be included in a Bayesian network analysis, and which are excluded from it. In contrast, we propose a method that scales well to tens of thousands of genes, allowing the inclusion of all genes for learning the network. Our new algorithm, called S-DAG, splits the large set of genes into small subsets, and learns a Bayesian network for each subset independently. Afterwards, the subnetworks are combined to one large network, considering the constraint that the structure of a Bayesian network must not contain cycles. Based on benchmark data, we show that our algorithm reconstructs large-scale networks with a higher quality than the very competitive MMHC algorithm. Additionally, we apply our algorithm on a <i>S. cerevisiae</i> and a <i>homo sapiens</i> data set. The latter data set contains over 50,000 genes, which is the largest unrestrictred Bayesian network reconstructed from data, so far. Results from both data sets show that the learned networks have a high biological relevance. KW - gene_expression KW - genetic_network KW - own_publication KW - structure_learning AU - Nägele, Andreas AU - Arndt, Holger AU - Dejori, Mathäus A2 - Tsakalides, A A2 - Skarlas, L PY - 2008/07/22/ ER -