A graphical model method for integrating multiple sources of genome-scale data

Abstract

Making effective use of multiple data sources is a major challenge in modern bioinformatics. Genome-wide data such as measures of transcription factor binding, gene expression, and sequence conservation, which are used to identify binding regions and genes that are important to major biological processes such as development and disease, can be difficult to use together due to the different biological meanings and statistical distributions of the heterogeneous data types, but each can provide valuable information for understanding the processes under study. Here we present methods for integrating multiple data sources to gain a more complete picture of gene regulation and expression. Our goal is to identify genes and cis-regulatory regions which play specific biological roles. We describe a graphical mixture model approach for data integration, examine the effect of using different model topologies, and discuss methods for evaluating the effectiveness of the models. Model fitting is computationally efficient and produces results which have clear biological and statistical interpretations. The Hedgehog and Dorsal signaling pathways in Drosophila, which are critical in embryonic development, are used as examples.

DNA methylation and childhood asthma in the inner city

“…To further explore the relationship between DNA methylation and gene expression, we performed an integrative analysis of DNA methylation and expression using LCMix 44 and identified 2484 methylation-expression pairs (posterior probability > 0.95 from the joint fit corresponding to a local false discovery rate <0.05), including RUNX3 and IL4 , the expression of which was related to asthma-associated DNA methylation marks (Fig 3, C , and see Table E10 in this article’s Online Repository at www.jacionline.org). Among the genes with the most significant methylation-expression relationship is an additional asthma-associated gene, ST2 , the receptor for IL-33.…”

“…44 LCMix requires a 1:1 relation between the input data sets, and therefore we limited our analysis to expression probes that had a methylation probe upstream within 3 kb of the transcription start site (23,848 genes). We then determined the number of components for the marginal fits to methylation and expression using integrated classification likelihood-Bayesian information criterion, 45 as suggested in Dvorkin et al 44 Using those component numbers, we then performed joint fit using the chained topology (methylation and expression), 2 hidden components, and the Pearson type VII family.…”

DNA Methylation and Childhood Asthma in the Inner-City

“…The finding that many of the inverse relationships are within gene bodies is in agreement with a recent publication that showed prominent hypomethylation of gene bodies with both inverse and direct correlation to gene expression in cancer 50 and demonstrates that the relationship of gene body methylation and expression is more complex than previously thought. 51 To further explore the relationship between DNA methylation and gene expression, we performed an integrative analysis of DNA methylation and expression using LCMix 44 and identified 2484 methylation-expression pairs (posterior probability > 0.95 from the joint fit corresponding to a local false discovery rate <0.05), including RUNX3 and IL4, the expression of which was related to asthma-associated DNA methylation marks (Fig 3, C, and see Table E10 in this article's Online Repository at www.jacionline.org). Among the genes with the most significant methylation-expression relationship is an additional asthmaassociated gene, ST2, the receptor for IL-33.…”

“…To integrate the expression and methylation data, we used LCMix, which was shown to be effective at identifying target genes in real and simulated data sets by combining information from disparate data sets. 44 LCMix requires a 1:1 relation between the input data sets, and therefore we limited our analysis to expression probes that had a methylation probe upstream within 3 kb of the transcription start site (23,848 genes). We then determined the number of components for the marginal fits to methylation and expression using integrated classification likelihood-Bayesian information criterion, 45 as suggested in Dvorkin et al 44 Using those component numbers, we then performed joint fit using the chained topology (methylation and expression), 2 hidden components, and the Pearson type VII family.…”

Semi-Supervised Learning Using Hierarchical Mixture Models: Gene Essentiality Case Study

“…However, even training data whose labels are too incomplete for supervised approaches can provide valuable information beyond unsupervised analysis. In this work, we describe the extension of the unsupervised methods first described in [25] to allow the use of any available positively labeled training data, even if limited, for semi-supervised learning. Alexandridis et al [26] describe an ad hoc method for semi-supervised mixture modeling with incomplete training data.…”