GO-PCA requires two types of inputs: First, a gene expression matrix (see Running GO-PCA), and second, a list of gene sets. These gene sets represent the prior knowledge used by GO-PCA, and form the basis for signature generation. Generally speaking, they should represent functional categories of potential relevance to the expression dataset analyzed. This section discusses different types of gene sets and how to obtain gene set files for use with GO-PCA.
In short, pre-generated GO-derived gene set files (and corresponding Gene
Ontology files in OBO format) are available for download (see below).
Alternatively, users can create and use their own, customized gene sets.
GO-PCA was originally designed to work with gene sets derived from Gene Ontology (GO) annotation data in the UniProt-GOA database. Each GO term gives rise to a particular gene set, which is defined as the set of all genes annotated with that term. However, not all such gene sets are necessarily useful for GO-PCA: There are many overly specific GO terms that are associated with very few genes, and there are some overly general GO terms that are associated with too many genes. Moreover, many GO annotations are uncertain, sometimes based solely on computational predictions. These issues need to be taken into consideration when deriving gene sets from GO annotation data.
Note
The hierarchical structure of the Gene Ontology also results in many
highly overlapping gene sets. However, this is anticipated by the GO-PCA
algorithm, which includes two different filtering steps designed to limit
the generation of redundant signatures. For this filtering to work as
intended, GO-PCA also needs to be given an “ontology file”, containing the
structure of the Gene Ontology (in OBO format).
While GO-derived gene sets represent a high-quality, all-purpose repository of prior knowledge, many GO-PCA analyses might benefit from alternative or additional gene sets that either serve to supplement the UniProt-GOA database, or represent more context-specific prior knowledge that a researcher might have acquired from previous experiments. GO-PCA will treat such custom gene sets in the same way as GO-derived gene sets, with the minor exception that certain certain filtering rules do not apply (see previous Note).
Up-to-date, GO-derived gene sets based on high-confidence GO evidence codes are available for download. For more details on the procedure used to derive these gene sets from GO annotation data, please see the included README.
Note that GO-PCA should be provided with both the gene set file (.tsv) and the matching Gene Ontology file (.obo), in order to be able to apply all filters (see previous Note). For more information on how to specify both gene sets and Gene Ontology when running GO-PCA from the command- line, see Running GO-PCA.
Custom GO-derived gene sets can be generated using the script gopca_extract_go_gene_sets.py.