Automatic comparison of metabolites names: impact of criteria thresholds

The growing number of stoichiometric reconstructions and models tends to change the model building process. Instead of creating a new model from scratch scientists can look at the earlier created relevant models to assess the opinion and consensus level of other modellers. Several initiatives have been performed to build consensus models for particular organisms following this approach. One of possible improvements in the model development taking into account earlier developed ones is automated comparison of models. That is enabled by the fact that models usually are in a computer readable format to be simulated. Still there are some problems like different ways of naming metabolites, models without formulae of metabolites, different approach in definition of compartments and other peculiarities of different research groups at different times. There are several software tools that offer reconciliation or mapping of metabolites in models to assess the similarity of models. It is computationally trivial to find metabolite pairs with identical names in two models. Still very often the comparison algorithms lack flexibility and some work should be done by manual curation of metabolite pairs to recognize that the difference is caused by symbols like brackets, quotes, apostrophes, spaces, upper/lower case letters or similar ones. The proposed approach suggests combination of automated comparison with manual curation: most of possible metabolite pairs are rejected by computer leaving just the most similar metabolite pairs for manual comparison. The elasticity in metabolite name comparison is introduced using Levenstein similarity ratio and Levenstein edit distance. Application of these criteria with different acceptance thresholds is analyzed comparing two models of Saccaromyces cerevisiae with 681 and 1063 metabolites. The results are compared with manually approved pairs of matching metabolites.