In this section, we compare two methods that can handle missing annotations in the labels: binary relevance and classifiers chains.
In making this decision, the curator takes into account information such as duplication events/orthology, sequence conservation, the presence of essential/active site residues, branch length, and genes having inconsistent experimental annotations (i.e. descendants with annotations, or missing annotations in well-characterized genes, that are most likely not compatible with the annotation).
Due to the missing annotation in the A. annua contig database a BLASTx search with an E-value cut-off of 0.01 was performed to translate the contigs into known protein homologues.
Genes and proteins in the network neighbourhood, in particular those with missing annotations, are not taken into account.
However, this approach has four drawbacks: (i) it can only score functional associations of overlapping gene/proteins sets; (ii) it disregards genes with missing annotations; (iii) it does not take into account the network structure of physical interactions between the gene/protein sets of interest and (iv) tissue-specific gene/protein set associations cannot be recognized.
These methods are easily understood and realized, but they are readily misled into making an error-prone prediction by spurious and missing annotations of proteins.
Deletions were not considered into analyses of gene turnover, since they cannot be distinguished from missing annotations.
This formalism tolerates missing annotations at intermediate levels as, in this case, it can aggregate lower level information into the first upper level where information is available.
For each metabolite, we verified the existing annotations and attempted to fill in missing annotations of the remaining four input types.
For 72 of the 266 genes with missing annotations, we could obtain annotations from version 7.0 (Additional file 2: Table S2) and one gene was marked as obsolete.
