Figure 1

Procedure for applying machine learning methods to metabolic pathway prediction. Data from curated pathway/genome databases (PGDBs) are gathered into a "gold standard" collection. Features are defined using biological knowledge, and their values are computed for all pathways in the gold standard. The resulting dataset is split into training and test sets. Training data are used to perform feature selection and parameter estimation for multiple predictor types. Test data are used to evaluate the predictors. The predictor which performs best on the test set will be applied to data from newly sequenced and annotated genomes to perform metabolic network reconstruction.