Example datasets

Ontologizer 3 provides an example dataset available to download from Zenodo. This dataset contains the necessary Gene Ontology (GO) file, along with GO association files (GAF) and simulated datasets for five commonly studied organisms.

Common name	Organism	GAF file
fly	Drosophila melanogaster	fb.gaf
human	Homo sapiens	goa_human.gaf
mouse	Mus musculus	mgi.gaf
rat	Rattus norvegicus	rgd.gaf
yeast	Saccharomyces cerevisiae	sgd.gaf

To try one of the datasets, launch Ontologizer and either download the GO and GAF files from within the application or load them from the bundle. Then load the corresponding population and study gene files, perform the desired analysis and explore the enriched terms.

Content

Each organism folder contains:

File	Contents
population_genes.txt	All protein-coding genes of the organism, one gene symbol per line.
study_genes.txt	A simulated study set drawn from the population, enriched for a small number of GO terms.
solution.tsv	The ground-truth causal terms used to construct the study set, with the specific genes drawn from each.

Construction

The study_genes.txt file follows the simulation procedure described in the Ontologizer 3 paper. GO terms were sampled at random and a fraction $\rho$ of each term's annotated genes was added to the study set, until the study set reached a target size. Unrelated noise genes were then added so that a target fraction $\eta$ of the study set is annotated to a sampled term. Each dataset uses $\rho = 0.4$ (40% of each sampled term's annotated genes are included in the study set) and $\eta = 0.5$ (half the study-set genes are annotated to a sampled term, half are noise), corresponding to a moderate-signal, moderate-noise regime.

The solution.tsv file is tab-separated with two columns, containing the GO term ID (or the label Noise for unrelated genes) and a comma-separated list of the genes drawn from that term and added to the study set:

GO:0035196  PUS10,NCBP2,AGO1,LIN28B,TNRC6C,...
GO:0030018  SYNPO2L,PDLIM7,SYNC,NEB,...
GO:0097553  PLCB4,TRPV3,ERO1A,...
...
Noise       CALR,SNX31,HCRTR2,PTPN22,OPN1MW3,CLEC16A,...

Try it

A useful exercise to build some inuition on the two methods is to run them on the same dataset and to compare the reported terms against solution.tsv. MGSA typically returns a much shorter list, with most of the reported terms appearing in solution.tsv. Fisher's exact test will also identify the causal terms, but among a longer list of redundant terms, typically ancestors of the causal terms, which are enriched as a consequence of the true-path rule. This precision difference is quantified in the manuscript's validation experiments.