• Molloy, Peter
• Anwar, Firoz
• Wilson, William
• Bauer, Denis

Abstract

Gene Ontology (GO) enrichment analysis can provide insight into the underlying biology function common to a list of interesting genes, e.g. differentially expressed (DE) genes. To determine whether a GO term is over/under-representated in this set, one needs to compare the frequency of this term in a set of genes considered baseline (background). The choice and size of the background can influence the identified terms and their significance, potentially resulting in misleading biological interpretation. This problem is specifically pronounced in situations where enrichment is performed in an already very distinctive subset of the genome, e.g. mitochondrial proteins.We therefore propose a new approach for selecting the background and pruning the GO term tree for the analysis of gene subsets with distinctive function. To compare our method against traditional approaches,we perform GO enrichment analysis using goseq [1] on 93 differentially expressed (DE) mitochondrial genes (mitochondrial protein compendium, MitoCarta [2]) between visceral and subcutaneous adipocyte cells in human (3 matched biological replicates) using different background and pruning approaches.Performing the GO analysis on the 93 DE genes using the full set of GO terms and all non differentially expressed genes in the genome as background (57,818) we obtain 161 significant terms. However, the most significant terms cover broad mitochondrial function. For example, the top three terms are the cellular compartment terms ‘intracellular’, ‘cytoplasm’, ‘mitochondrion’ and are hence not capturing the regulatory difference of the mitochondrial genome between the different fat types. To focus on mitochondrial function, we limit the background to only cover MitoCarta genes (1041), however this results in no significant terms being identified. We know that mitochondria play a crucial role in adipogenesis and hence the two tissues are likely to have different function. This highlights the need for constructing a purpose-built background.Starting from the GO terms associated with MitoCarta genes, we include all genes that share these GO terms as the background, which reveals 163 enriched terms. Extending this approach, we also remove the GO terms not associated with MitoCarta genes as they are not able to reach significance and impede the multiple testing correction. This final approach of tailored background and pruned GO tree results in 226 enriched. These new terms where particularly enriched in GO categories containing less than 500 genes. These considerably more specific GO terms relate to relevant biological processes such as beta-oxidation of fatty acids and acetyl-CoA metabolism. This finding suggests that by constructing a purpose-built background set of genes and a pruned GO tree we are able to identify some fundamental difference in the catabolism of fats between visceral and subcutaneous adipocytes.

Topics

• impedance spectroscopy
• laser emission spectroscopy