Application of string similarity ratio and edit distance in automatic metabolite reconciliation comparing reconstructions and models

Increasing numbers of biochemical network models become available and reuse of these models is becoming more common. As a consequence, tools to compare models are needed. Comparison can be difficult because model builders often use different standards during reconstruction, metabolite formulas are not always indicated, IDs and names of metabolites are different, and models are stored in different formats (SBML, COBRA and others). Herein, a model comparison algorithm for SBML and COBRA format models is presented, called ModeRator. Precondition for correct matching of reactions is the comparison of the participating metabolites. ModeRator is based on the comparison of metabolite names as text strings. An automatic three level filtering approach is implemented in the software, which rejects pairs of potentially equal metabolites and builds an opinion about metabolite pairs with high similarity in metabolite names. ModeRator was applied to two test cases, comparing two models of each, E.coli and S.cerevisiae. Matches of the automatic mapping were manually inspected and compared with the automatic predictions. Automatic metabolite mapping of E.coli models (1314 and 1704 metabolites) comparing only identifiers revealed a high number of accordant metabolites. Both models originate from the same source (BioCyc database). No significant difference between automatic mapping and manual curation are observed. For the comparison of two S.cerevisiae models (679 and 1061 metabolites), three level filtration by metabolite name is used. The discrepancy between manual curated predictions and ModeRator predictions was 7%.