Additional file 3 of GeneWalk identifies relevant gene functions for a biological context using network representation learning

Additional file 3. Gene function benchmark ground truth and model predictions for qki and JQ1 contexts. The sheets named after the publications used to generate the ground truth (SciRep2016Darbelli, MolCell2016Winter, CancerCell2012Sanda_abstract and CancerRes2018Sharma_abstract) contain all relevant genes, biological context terms and text sections extracted from the respective publications (all columns are further described in detail in Materials and Methods section: Context-specific gene-function benchmarks). The remaining two sheets (Benchmark_qki and Benchmark_JQ1) contain all the DE genes and their GO annotations that corresponded to the benchmark genes. For GW (INDRA) the connected GO terms originating from INDRA through its text mining are also listed. The column bm_truth indicates the binary ground truth: TRUE if any context term from the ground truth publications matched the GO term through regular expression and FALSE otherwise (see Methods section: Context-specific gene-function benchmarks for details). Also shown are the FDR adjusted p-values (padj) values for each method and the corresponding binarized values (TRUE if padj< 0.1, FALSE otherwise). These were used to compare the performances of each method against the ground truth values.