Limits...
Comparison of multiplex meta analysis techniques for understanding the acute rejection of solid organ transplants.

Morgan AA, Khatri P, Jones RH, Sarwal MM, Butte AJ - BMC Bioinformatics (2010)

Bottom Line: We demonstrate that a commonly used parametric effect size estimate approach and a commonly used non-parametric method give very different results in prioritizing genes.Different methods of multiplex analysis can give substantially different results.The method which is best for any given application will likely depend on the particular domain, and it remains for future work to see if any one method is consistently better at identifying important biological signal across gene expression experiments.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Pediatrics, Stanford University School of Medicine, Stanford, CA 94305, USA.

ABSTRACT

Background: Combining the results of studies using highly parallelized measurements of gene expression such as microarrays and RNAseq offer unique challenges in meta analysis. Motivated by a need for a deeper understanding of organ transplant rejection, we combine the data from five separate studies to compare acute rejection versus stability after solid organ transplantation, and use this data to examine approaches to multiplex meta analysis.

Results: We demonstrate that a commonly used parametric effect size estimate approach and a commonly used non-parametric method give very different results in prioritizing genes. The parametric method providing a meta effect estimate was superior at ranking genes based on our gold-standard of identifying immune response genes in the transplant rejection datasets.

Conclusion: Different methods of multiplex analysis can give substantially different results. The method which is best for any given application will likely depend on the particular domain, and it remains for future work to see if any one method is consistently better at identifying important biological signal across gene expression experiments.

Show MeSH

Related in: MedlinePlus

ROC curves showing the relative ability of the meta fold-changes and the meta p-values to prioritize genes associated with "DEFENSE RESPONSE" as identified in the Molecular Signatures Database. Note that the overall area under the curve is superior for the prioritization of genes via meta fold-change from the inverse variance meta effect size estimate (MetaEffect), and this method is better at identifying genes at the very top of the list, as indicated by the dominance of the ROC curve at the far left, indicating very strong enrichment (higher true positive rate), while the number of false positives are still relatively low. This result suggests that the meta fold-changes provide a superior method for identifying the key genes in the undesirable immune response in acute organ rejection.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2967747&req=5

Figure 3: ROC curves showing the relative ability of the meta fold-changes and the meta p-values to prioritize genes associated with "DEFENSE RESPONSE" as identified in the Molecular Signatures Database. Note that the overall area under the curve is superior for the prioritization of genes via meta fold-change from the inverse variance meta effect size estimate (MetaEffect), and this method is better at identifying genes at the very top of the list, as indicated by the dominance of the ROC curve at the far left, indicating very strong enrichment (higher true positive rate), while the number of false positives are still relatively low. This result suggests that the meta fold-changes provide a superior method for identifying the key genes in the undesirable immune response in acute organ rejection.

Mentions: To attempt to identify which method is better at identifying biologically relevant genes in this data set, we use the Molecular Signatures Database [10] to identify the lists of genes associated with "Immune Response" and "Defense Response". A gene expression meta analysis approach that is being used to probe the undesirable immune response in transplant rejection should prioritize these genes more highly. We can compare the prioritization provided by Fisher's method p-values (using the lower p-values as indicating higher priority) and the meta fold-change provided by the inverse variance method to select these import immune function genes. These results are shown in Figures 2 and 3. It can be seen that the ranking provided by the effects size estimate from the inverse variance method, the meta fold-change, is superior to that provided by Fisher's method, the meta p-values. The Fisher's method also does not differentiate itself from the prioritization provided by the individual constituent datasets being combined.


Comparison of multiplex meta analysis techniques for understanding the acute rejection of solid organ transplants.

Morgan AA, Khatri P, Jones RH, Sarwal MM, Butte AJ - BMC Bioinformatics (2010)

ROC curves showing the relative ability of the meta fold-changes and the meta p-values to prioritize genes associated with "DEFENSE RESPONSE" as identified in the Molecular Signatures Database. Note that the overall area under the curve is superior for the prioritization of genes via meta fold-change from the inverse variance meta effect size estimate (MetaEffect), and this method is better at identifying genes at the very top of the list, as indicated by the dominance of the ROC curve at the far left, indicating very strong enrichment (higher true positive rate), while the number of false positives are still relatively low. This result suggests that the meta fold-changes provide a superior method for identifying the key genes in the undesirable immune response in acute organ rejection.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2967747&req=5

Figure 3: ROC curves showing the relative ability of the meta fold-changes and the meta p-values to prioritize genes associated with "DEFENSE RESPONSE" as identified in the Molecular Signatures Database. Note that the overall area under the curve is superior for the prioritization of genes via meta fold-change from the inverse variance meta effect size estimate (MetaEffect), and this method is better at identifying genes at the very top of the list, as indicated by the dominance of the ROC curve at the far left, indicating very strong enrichment (higher true positive rate), while the number of false positives are still relatively low. This result suggests that the meta fold-changes provide a superior method for identifying the key genes in the undesirable immune response in acute organ rejection.
Mentions: To attempt to identify which method is better at identifying biologically relevant genes in this data set, we use the Molecular Signatures Database [10] to identify the lists of genes associated with "Immune Response" and "Defense Response". A gene expression meta analysis approach that is being used to probe the undesirable immune response in transplant rejection should prioritize these genes more highly. We can compare the prioritization provided by Fisher's method p-values (using the lower p-values as indicating higher priority) and the meta fold-change provided by the inverse variance method to select these import immune function genes. These results are shown in Figures 2 and 3. It can be seen that the ranking provided by the effects size estimate from the inverse variance method, the meta fold-change, is superior to that provided by Fisher's method, the meta p-values. The Fisher's method also does not differentiate itself from the prioritization provided by the individual constituent datasets being combined.

Bottom Line: We demonstrate that a commonly used parametric effect size estimate approach and a commonly used non-parametric method give very different results in prioritizing genes.Different methods of multiplex analysis can give substantially different results.The method which is best for any given application will likely depend on the particular domain, and it remains for future work to see if any one method is consistently better at identifying important biological signal across gene expression experiments.

View Article: PubMed Central - HTML - PubMed

Affiliation: Department of Pediatrics, Stanford University School of Medicine, Stanford, CA 94305, USA.

ABSTRACT

Background: Combining the results of studies using highly parallelized measurements of gene expression such as microarrays and RNAseq offer unique challenges in meta analysis. Motivated by a need for a deeper understanding of organ transplant rejection, we combine the data from five separate studies to compare acute rejection versus stability after solid organ transplantation, and use this data to examine approaches to multiplex meta analysis.

Results: We demonstrate that a commonly used parametric effect size estimate approach and a commonly used non-parametric method give very different results in prioritizing genes. The parametric method providing a meta effect estimate was superior at ranking genes based on our gold-standard of identifying immune response genes in the transplant rejection datasets.

Conclusion: Different methods of multiplex analysis can give substantially different results. The method which is best for any given application will likely depend on the particular domain, and it remains for future work to see if any one method is consistently better at identifying important biological signal across gene expression experiments.

Show MeSH
Related in: MedlinePlus