|
Portuguese researchers have developed a technique for classifying genomic-wide metabolic reactions, which they suggest will open up a new approach to diversity analysis of metabolic reactions and comparison of metabolic pathways as well as being generally compatible with the conventional "EC" classification of enzymes. To Diogo Latino and Joćo Aires-de-Sousa of the New University of Lisbon the EC system is rather limited when it comes to reconstructing a metabolic pathway from genomic data, not least because the system is not without ambiguities. The functions of enzymes are usually described using an EC number, which also identifies the reactions, enzymes and their genes, which ties metabolic and genomic information. It is the rules for assigning an EC number that vary between different types of reactions so that their direct use in studying all the diverse metabolic reactions at the genomic level (reactomics, to give it the right buzzword) is limited, the researchers say. They have built on the work of Johann Gasteiger of the University of Erlangen, Germany and colleagues in using Kohonen neural networks to organise data by creating a space into which chemical reactions can be placed in a meaningful way. As such, Latino and Aires-de-Sousa have now mapped the reactome into a self-organizing map (SOM), which can be used to classify metabolic reactions, to assign EC numbers based on the molecular structures of the substrates and products, and moreover to define metabolic reaction space. Gasteiger and colleagues devised a method of numerically labelling a chemical bond based on seven properties relevant to its reactivity and used their neural network to map bonds of interest into a reactivity space. The Portuguese team similarly "trained" a 15 X 15 neural network (their SOM) with various bonds randomly selected from the milieu of metabolic compounds, with each bond's reactivity dictated by the seven physicochemical properties described by Gasteiger. Bonds close to each other on the resulting SOM are then revealed as being similar, while those far apart are more distantly related, if at all. When the bonds of a given compound are presented to the trained SOM, the team reveals the pattern of activated places as a MOLMAP (a molecular map of atom-level properties) for that compound. By next looking at the differences between the MOLMAP of the products and the MOLMAP of the reactants the team was able to extract descriptors of the specific metabolic reactions. The differences represent a reaction "fingerprint" and this information can be used to classify a vast range of metabolic reactions as well as predict the properties of "new" or unknown reactions by feeding the atom-level details of substrate and product into the neural network. The team applied the MOLMAP approach to 3468 reactions from the KEGG LIGAND database of enzymatic reactions. This sub-set contains reactions catalyzed by 1105 oxidoreductases, 1098 transferases, 622 hydrolases, 335 lyases, 172 isomerases, and 136 ligases. The analysis revealed relationships between reactions that are not apparent from their simplistic EC numbers. Moreover, the approach can identify problematic EC classifications. de Sousa explains the potential impact of his work further, "Genes related to enzymes are usually annotated with EC numbers," he told SpectroscopyNOW, "In order to assign genes to specific reactions in metabolic pathways, enzyme scientists can predict the EC numbers for the reactions and retrieve genes annotated with the same EC numbers in reference databases." 'Enzyme function' usually refers to the chemical reactions that the enzyme catalyses. An enzyme scientist can query the system with a reaction to find out if a similar reaction is known in the reactome. Specifically, if an enzyme scientist wants to identify an enzyme with a similar function to another enzyme, then submitting the MOLMAP of the corresponding chemical reaction to the system will provide a list of the most similar metabolic reactions known, and the enzymes that catalyse them. Organic chemists could also data mine the reactome using this system to find possible biocatalysts for an interesting organic reaction. "This approach shows a new way to explore the diversity of metabolic reactions," he adds, "An important task in the drug discovery process is the assessment of how a compound is metabolised. A physicochemical parameter-based approach for reaction classification has a potential for advancing to reactivity classification, which could eventually suggest reactions involved in the metabolisation of a compound. " Related links:
|
Bringing colour to reactomics |
