Pathway analysis for genomics data
- 1. Pathway analysis of genomic data: concepts, methods, and prospects for future development 1 By: Sakshi Jha
- 2. Content s • Introduction • What do we get out of pathway analysis • Selecting an overall study design • Fundamental concepts about biological pathways and networks • Preparing data for association testing • Analytical methods to detect pathway–phenotype relationships • Post-analysis considerations • Future developments in genomic data analysis • Pathways and networks: bridging multi-omics data • References 2
- 3. Introduction • In the last decade, pathway analysis has become the first choice for extracting and explaining the underlying biology for high throughput molecular measurements. • Today, virtually every bioinformatics study looks for statistically significant pathways as either biological interpretation or validation of computationally derived results. • Genome-wide data sets are increasingly viewed as foundations for discovering pathways and networks relevant to phenotypes. 3
- 4. What do we get out of pathway analysis? • In-depth and contextualized findings to help • understand the mechanisms of disease in question • Identification of genes and proteins associated with the etiology of a specific disease • Prediction of drug targets • Understand how to intervene therapeutically in disease processes • Conduct targeted literature searches. 4
- 5. 5
- 6. Selecting an overall study design 6 Candidate Pathway Analysis is hypothesis driven: pathways are preselected based on prior knowledge and insight. • Although the number of candidate pathways may vary with study goals. • Candidate pathway analysis most appropriate where computational resources are limited and where specific pathways are of a priori interest. GENOME WIDE PATHWAY ANALYSIS, interrogates a complete genomic data set through pathways representing an extensive range of biology. • Methods limited to GWPA have been used on data sets with only1000 genes. • This approach can more readily detect unexpected relationships. • GWPA is computationally intensive • Requiring more stringent corrections for multiple comparisons and making procedures, such as imputation, more challenging.
- 7. 7
- 8. Fundamental concepts about biological pathways and networks 8
- 9. Obtaining input genomic and pathway annotation data 9 o Pathway analyses can utilize raw genotype data for individual subjects or a list of p-values relating genes or SNPs To a phenotype. o In parallel, a pathway analysis is only as good as the functional information underlying its pathway definitions. o Investigators should consider their resources and study goals when selecting the most appropriate genomic data source. o Prominent pathway annotation databases exhibit diverse features. o The ideal choice of database depends on several variables and their impact on study goals
- 10. 10
- 11. PREPARING DATA FOR ASSOCIATION TESTING 11
- 12. Pathway size • Most pathway analyses place constraints on pathway size: • Small pathways can exhibit false positive associations because of large single-gene or single-SNP effects, whereas large pathways are more likely to show association by chance alone. • The most common minimum threshold for pathway size appears to be 10 genes. • Frequently used maximum thresholds for pathway size include 100 and 200 genes. • Often, this threshold may exclude highly specific and potentially informative functional sets, including those involving protein complexes and DNA sequence motifs. • Overall, investigators should consider their study goals when applying such thresholds and should evaluate results in that context. 12
- 13. Pathway overlap • Genes and their products typically act in multiple pathways, and each role is potentially important to a disease or treatment mechanism. As a result, analyses can expect to have some degree of pathway overlap. • High pathway overlap can obscure the true source of an association signal; this problem can exist with any pathway analysis, Gene Ontology (GO) annotations are particularly susceptible because of the large, hierarchical structure of the database. • Pathway overlap can also be addressed during post-analysis to prioritize related pathways for further exploration. • Extant strategies include hierarchical clustering in a study of breast cancer, overlap-based network creation in the visualization tool ‘Enrichment Map’ and the listing of overlapping pathways alongside results in the analytical software PARIS. 13
- 14. Assigning data elements to genes • Genomic data have historically been integrated into pathways by mapping assayed elements to genes. • SNP based genotyping arrays, this is not straightforward because many array SNPs are not located in known coding or regulatory regions. • Alternatively, each unmapped SNP can be assigned to its nearest gene. • Evolving theories suggest that sequences are not associated to genes based on closest proximity, and may not even be solely associated to one gene. • Hence, many studies assign unmapped SNPs to all genes within a distance window, ranging from 10 kb to 500 kb. • SNPs that map to multiple genes in the same pathway can yield spurious pathway association. This issue is particularly important for MHC/HLA genes. 14
- 15. Calculating gene significance and accounting for LD • Most pathway analysis tools utilize one association signal per gene. • Whereas expression arrays yield a single P-value for each gene, SNP arrays include multiple signals per gene, some of which are correlated. • some studies use the minimum SNP-level P-value within a gene as the operative signal; • however, this approach will not detect additive effects among SNPs with moderate individual association. • For methods that combine SNP level signals, including those based on the truncated product method , LD must be accounted for to prevent highly correlated SNPs from biasing gene- level significance. 15
- 16. • Strategies to accomplish this include discarding SNPs that depart from LD at a pre-set threshold and adapting principal component analysis to extract the most independent signals within a gene. 16
- 17. Analytical methods to detect pathway–phenotype relationships • Attempting to better understand the molecular basis of a phenotype, studies often identify the pathways to which the genes that may be linked to the SNPs belong. • However, to establish the association of the phenotype with a pathway, one needs to show that phenotype associated SNPs tend to fall within that pathway more than expected by chance. • Following data processing, analytical methods can be applied to test for significant pathway– phenotype relationships. Prominent examples of pathway-based analytical tools and their salient features are shown: 17
- 18. 18
- 19. Post-analysis considerations • Following pathway analysis, appropriate data reporting and interpretation are imperative. • Currently, bias introduced by gene size is less commonly addressed than is bias from pathway size. • In particular, large genes containing many SNPs are more likely to contain significant SNPs by chance alone; for analyses, this can favor pathways containing large genes. • Analytical tools that use permutations naturally control for gene size by comparing the actual association data with the distribution of association statistics generated from randomly permuted data sets. • Other approaches allow users to restrict analysis to a subset of the most significant SNPs in each gene: for large genes, this may eliminate some spuriously associated SNPs and thus limit their impact on the pathway analysis. 19
- 20. • Other sources of bias that should be addressed include the capacity for strongly associated markers to drive pathway association and the possible effects of SNPs being assigned to multiple genes. • Fundamentally, these approaches to bias are best complemented by replication of pathway analysis findings in independent data sets. • key choices in pathway analyses will limit major contributors of variance across studies and will guide investigators in selecting approaches that fit their study goals. 20
- 21. Future developments in genomic data analysis • Development of methods and tools related to pathway analysis is ongoing and dynamic. • In particular, because pathways are of broad interest, targeted adaptations to their associated databases would expand their utility for investigators from a variety of backgrounds. • These adaptations might include simpler search and download mechanisms, consistency in pathway names and classifications, and methods for describing pathway overlap. • In addition, a universal format for annotation files might encourage interoperability among analytical tools, allowing investigators flexibility to match precisely their databases and statistical methods of choice. • Two recent trends among databases are also promising Specialized disease databases, such as AlzGene and the UCSC Cancer Genomics Browser. 21
- 22. • Meanwhile, complementary methods can extend the biological reach of pathway- based results. • pathway analysis platforms, computational efficiency will be highly valued, given the impressive granularity of next-generation sequencing data. 22
- 23. PATHWAYS AND NETWORKS: BRIDGING MULTI-OMICS DATA • In the coming years, we anticipate that pathways and network will assume a farther-reaching role in view of the need to integrate multi-omics data through systems biology approaches. • A variety of large-scale strategies are being used to study complex diseases, including genomic, transcriptomic, proteomic and metabolomic approaches, and data from all of these sources can be analyzed through pathways and networks representing coordinated functions and relationships. • As such, the role of pathways and networks as the hub for this integration will be vital in the years to come. 23
- 24. Reference • (Ramanan, Shen et al. 2012) Ramanan,V. K., L. Shen, J. H. Moore and A. J. Saykin (2012). "Pathway analysis of genomic data: concepts, methods, and prospects for future development."TRENDS in Genetics 28(7): 323-332. https://doi.org/10.1016/j.tig.2012.03.004 24
- 25. 25 Thank You