How Chemoinformatics can save the Human Metabolome Project

Recently I attended the 1st Australasian Symposium on Metabolomics and I was lucky enough to listen the plenary talk given by David Wishart. David leads Human Metabolome Project(HMP), a metabolomics counterpart of the human genome project, mainly funded by Genome Canada. Human Metabolome Project started in January 2005 with a mandate to identify, quantify, catalog and store all metabolites that can potentially be found in human tissues and biofluids at concentrations greater than one micromolar. In last fours years project has made substantial progress, with their concerted efforts David and his colleagues have cataloged several databases such as ToxDB, DrugMet, FooDB, DrugBank and most notably HMDB or Human Metabolome Database. No doubt HMDB is the most comprehensive collection of human metabolite data, but at the same time it is very much incomplete.

In past, HMDB has attracted several criticisms, particularly from the researcher in metabolomics community, regarding the coverage and the potential implications of HMDB. Jeremy Nicholson from the Imperial College London who suggested that HMDB is just a list of detectable metabolites, although a useful list. Apart from being incomplete, there are dramatic variations in metabolites profile because metabolic profile depends on several factors including age, sex, diet, ethnicity, fitness, and many more. To catalog the human metabolome, David and his group curated thousands of books, journal articles, and electronic databases. Further they used several metabolic profiling techniques such as NMR, GC-MS, and LC-MS. Surprisingly these analytical techniques failed to report several metabolites which were already known. Each of these techniques have pros & cons, and unfortunately there is no universal way to profile each metabolite. For example NMR based approach can not detect the inorganic ions while GC-MS based method is less sensitive. To this end a chemoinformatics based approach seems a reasonable solution to the problem of metabolite coverage, either a chemical transformation based top-down approach or a substructure/fragment based bottom-up approach can help to discover the complete human metabolite chemical space. In first case if there are 20,000 known metabolites then based on our knowledge about chemical transformations one can predict up to 2,00,000 metabolites and their spectra (MS, NMR etc) using available chemoinformatics tools. In second approach which exploits power of neural networks and genetic algorithms to explore metabolite space using already known metabolite substructures/fragments in a combinatorial fashion and predicts the spectra for them. In either case, subsequent matching of predicted spectra of potential metabolites with observed spectra of the tissue or body fluid sample can establish the newly identified metabolites. This also reminds me a article written by Lincoln D. Stein How Perl Saved the Human Genome Project, may be Chemoinformatics can do same job for Human Metabolome Project.