Limits...
Gene-set Analysis with CGI Information for Differential DNA Methylation Profiling.

Chang CW, Lu TP, She CX, Feng YC, Hsiao CK - Sci Rep (2016)

Bottom Line: Here we aimed to include both pathway information and CGI status to rank competing gene-sets and identify among them the genes most likely contributing to DNA methylation changes.Results show that, based on probabilities, the importance of pathways and genes can be determined.The findings confirm that some of these genes are cancer-related and may hold the potential to be targeted in drug development.

View Article: PubMed Central - PubMed

Affiliation: Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University, Taipei 10055, Taiwan.

ABSTRACT
DNA methylation is a well-established epigenetic biomarker for many diseases. Studying the relationships among a group of genes and their methylations may help to unravel the etiology of diseases. Since CpG-islands (CGIs) play a crucial role in the regulation of transcription during methylation, including them in the analysis may provide further information in understanding the pathogenesis of cancers. Such CGI information, however, has usually been overlooked in existing gene-set analyses. Here we aimed to include both pathway information and CGI status to rank competing gene-sets and identify among them the genes most likely contributing to DNA methylation changes. To accomplish this, we devised a Bayesian model for matched case-control studies with parameters for CGI status and pathway associations, while incorporating intra-gene-set information. Three cancer studies with candidate pathways were analyzed to illustrate this approach. The strength of association for each candidate pathway and the influence of each gene were evaluated. Results show that, based on probabilities, the importance of pathways and genes can be determined. The findings confirm that some of these genes are cancer-related and may hold the potential to be targeted in drug development.

No MeSH data available.


Related in: MedlinePlus

DNAm and CGI status.(A) Boxplots of differences in DNAm between matched pairs of ovarian cancer patients (cases) and normal controls for 100 randomly selected probes. The 76 red boxplots are for probes in CGI region; while the 24 black boxplots are for probes not in CGI. Probes in CGI tend to have larger variation in θij, indicating a larger degree of variability in DNAm between cases and controls. (B) Correlation plots of θij. The upper left panel contains correlations of θij from probes not in CGI; while the lower right panel is for probes in CGI. The correlation in each panel is larger, as compared to the correlations in the other two blocks, indicating a greater degree of similarity in θij, the differences in DNAm.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4836301&req=5

f2: DNAm and CGI status.(A) Boxplots of differences in DNAm between matched pairs of ovarian cancer patients (cases) and normal controls for 100 randomly selected probes. The 76 red boxplots are for probes in CGI region; while the 24 black boxplots are for probes not in CGI. Probes in CGI tend to have larger variation in θij, indicating a larger degree of variability in DNAm between cases and controls. (B) Correlation plots of θij. The upper left panel contains correlations of θij from probes not in CGI; while the lower right panel is for probes in CGI. The correlation in each panel is larger, as compared to the correlations in the other two blocks, indicating a greater degree of similarity in θij, the differences in DNAm.

Mentions: To examine by figures whether DNA methylation varies according to CpG island status, boxplots of the pair-wise differences in DNAm θij from 100 randomly selected probes were constructed (Fig. 2A) for the UKOPS study. This figure included 76 probes located in CGIs (called CGI probes) and 24 located outside (called non-CGI probes). It can be readily observed that the differences in DNAm show larger variability when the probes are located in CGIs (colored in red); while the probes located outside of CGIs (colored in black) tend to have smaller dispersion. In fact, among the original 27,578 probes, the means of the θij were −1.1 × 10−3 and −5.8 × 10−4, respectively (p < 1 × 10−7), supporting the assumption that CGI probes and non-CGI probes are not homogeneous. Moreover, probes of the same CGI status tend to have similar values of θij. Figure 2B shows the correlation between non-CGI probes and the correlation between CGI probes. A clear pattern emerges; probes are more alike if they are of the same CGI status. Both Fig. 2A,B support the assumption that the effect of CGI status βj for probe j can be assumed to come from one of two distributions, depending on the CpG island status of the probe. Similarly, the pattern of heterogeneous variation is apparent in the differences in DNAm among the 32 pairs of the lung cancer data. The boxplots of θij from 200 randomly selected probes in CGIs and from 200 not in CGIs show different degrees of variability, as seen in Supplementary Fig. S1a,b. For the NGS methylation data, the number of pairs is only 16, and thus no boxplot is produced.


Gene-set Analysis with CGI Information for Differential DNA Methylation Profiling.

Chang CW, Lu TP, She CX, Feng YC, Hsiao CK - Sci Rep (2016)

DNAm and CGI status.(A) Boxplots of differences in DNAm between matched pairs of ovarian cancer patients (cases) and normal controls for 100 randomly selected probes. The 76 red boxplots are for probes in CGI region; while the 24 black boxplots are for probes not in CGI. Probes in CGI tend to have larger variation in θij, indicating a larger degree of variability in DNAm between cases and controls. (B) Correlation plots of θij. The upper left panel contains correlations of θij from probes not in CGI; while the lower right panel is for probes in CGI. The correlation in each panel is larger, as compared to the correlations in the other two blocks, indicating a greater degree of similarity in θij, the differences in DNAm.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4836301&req=5

f2: DNAm and CGI status.(A) Boxplots of differences in DNAm between matched pairs of ovarian cancer patients (cases) and normal controls for 100 randomly selected probes. The 76 red boxplots are for probes in CGI region; while the 24 black boxplots are for probes not in CGI. Probes in CGI tend to have larger variation in θij, indicating a larger degree of variability in DNAm between cases and controls. (B) Correlation plots of θij. The upper left panel contains correlations of θij from probes not in CGI; while the lower right panel is for probes in CGI. The correlation in each panel is larger, as compared to the correlations in the other two blocks, indicating a greater degree of similarity in θij, the differences in DNAm.
Mentions: To examine by figures whether DNA methylation varies according to CpG island status, boxplots of the pair-wise differences in DNAm θij from 100 randomly selected probes were constructed (Fig. 2A) for the UKOPS study. This figure included 76 probes located in CGIs (called CGI probes) and 24 located outside (called non-CGI probes). It can be readily observed that the differences in DNAm show larger variability when the probes are located in CGIs (colored in red); while the probes located outside of CGIs (colored in black) tend to have smaller dispersion. In fact, among the original 27,578 probes, the means of the θij were −1.1 × 10−3 and −5.8 × 10−4, respectively (p < 1 × 10−7), supporting the assumption that CGI probes and non-CGI probes are not homogeneous. Moreover, probes of the same CGI status tend to have similar values of θij. Figure 2B shows the correlation between non-CGI probes and the correlation between CGI probes. A clear pattern emerges; probes are more alike if they are of the same CGI status. Both Fig. 2A,B support the assumption that the effect of CGI status βj for probe j can be assumed to come from one of two distributions, depending on the CpG island status of the probe. Similarly, the pattern of heterogeneous variation is apparent in the differences in DNAm among the 32 pairs of the lung cancer data. The boxplots of θij from 200 randomly selected probes in CGIs and from 200 not in CGIs show different degrees of variability, as seen in Supplementary Fig. S1a,b. For the NGS methylation data, the number of pairs is only 16, and thus no boxplot is produced.

Bottom Line: Here we aimed to include both pathway information and CGI status to rank competing gene-sets and identify among them the genes most likely contributing to DNA methylation changes.Results show that, based on probabilities, the importance of pathways and genes can be determined.The findings confirm that some of these genes are cancer-related and may hold the potential to be targeted in drug development.

View Article: PubMed Central - PubMed

Affiliation: Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University, Taipei 10055, Taiwan.

ABSTRACT
DNA methylation is a well-established epigenetic biomarker for many diseases. Studying the relationships among a group of genes and their methylations may help to unravel the etiology of diseases. Since CpG-islands (CGIs) play a crucial role in the regulation of transcription during methylation, including them in the analysis may provide further information in understanding the pathogenesis of cancers. Such CGI information, however, has usually been overlooked in existing gene-set analyses. Here we aimed to include both pathway information and CGI status to rank competing gene-sets and identify among them the genes most likely contributing to DNA methylation changes. To accomplish this, we devised a Bayesian model for matched case-control studies with parameters for CGI status and pathway associations, while incorporating intra-gene-set information. Three cancer studies with candidate pathways were analyzed to illustrate this approach. The strength of association for each candidate pathway and the influence of each gene were evaluated. Results show that, based on probabilities, the importance of pathways and genes can be determined. The findings confirm that some of these genes are cancer-related and may hold the potential to be targeted in drug development.

No MeSH data available.


Related in: MedlinePlus