Computational prediction of human metabolic pathways from the complete human genome

Abstract

Background: We present a computational pathway analysis of the human genome that assigns enzymes encoded therein to predicted metabolic pathways. Pathway assignments place genes in their larger biological context, and are a necessary first step toward quantitative modeling of metabolism.

Results: Our analysis assigns 2,709 human enzymes to 896 bioreactions; 622 of the enzymes are assigned roles in 135 predicted metabolic pathways. The predicted pathways closely match the known nutritional requirements of humans. This analysis identifies probable omissions in the human genome annotation in the form of 203 pathway holes (missing enzymes within the predicted pathways). We have identified putative genes to fill 25 of these holes. The predicted human metabolic map is described by a Pathway/Genome Database called HumanCyc, which is available at http://HumanCyc.org/. We describe the generation of HumanCyc, and present an analysis of the human metabolic map. For example, we compare the predicted human metabolic pathway complement to the pathways of Escherichia coli and Arabidopsis thaliana and identify 35 pathways that are shared among all three organisms.

Conclusions: Our analysis elucidates a significant portion of the human metabolic map, and also indicates probable unidentified genes in the genome. HumanCyc provides a genome-based view of human nutrition that associates the essential dietary requirements of humans with a set of metabolic pathways whose existence is supported by the human genome. The database places many human genes in a pathway context, thereby facilitating analysis of gene expression, proteomics, and metabolomics datasets through a publicly available online tool called the Omics Viewer.

Figure 1

A typical description of a gene product's function in Ensembl. This example aims to communicate to the reader exactly what information was obtained from Ensembl; it shows multiple functions, synonyms and EC numbers, as well as a Swiss-Prot accession number, all in one line of text. A Perl script was developed to parse these descriptions and extract the relevant information.

Figure 2

Predicted HumanCyc pathway for arginine degradation. The computer icon in the upper-right corner indicates this pathway was predicted computationally. Neither enzyme names nor gene names are drawn adjacent to the first three reactions of this pathway to indicate that these steps are pathway holes, meaning no enzyme has been identified for these steps in the human genome. The graphic at the bottom indicates the positions of genes within this pathways on the human chromosomes. Moving the mouse over a gene in the webpage for this diagram will identify the gene and the chromosome.

Figure 3

Curated HumanCyc pathway for oxidative ethanol degradation. This pathway was not predicted by PathoLogic, but was entered into HumanCyc as part of our subsequent literature curation effort. The flask icon in the upper-right corner indicates this pathway is supported by experimental evidence. The complete comment for this pathway is available at [38]

Figure 4

Numbers of pathways, including superpathways, shared by the three PGDBs HumanCyc (H. sapiens), EcoCyc (E. coli), and AraCyc (A. thaliana). The numbers outside the circles represent the total number of pathways in the corresponding PGDB. The numbers inside the intersecting areas represent the number of pathways that fall into each area. For example, there are 55 pathways in common between HumanCyc and EcoCyc (20 + 35). AraCyc contains 177 total pathways: 76 that are unique to A. thaliana, and 101 that are shared with other organisms.