Table 5.

Experimental validation of novel predictions: predicted (pred.) gene set refers to the set of novel (out-of-HGv1) target genes of a hormone predicted by our BioEmbedS* models (at default SVM probability cutoffs; see Section 4), and we compute its overlap with an experimentally derived reference (ref.) set of hormone-responsive differentially expressed genes (DEGs, including both up/down-regulated genes) already available for two well-studied hormones via gene expression omnibus (GEO) and Enrichr resources

Hormone	Ref. geneset name (data source and identifiers)	Ref. experiment description (Condition 1 versus 2 in DEG analysis)	Ref. geneset size	Pred. geneset size	Overlap size (overlap P-value)
Insulin	Insulin Receptor Associates Promoters GSE107336 (Hancock et al., 2019) (GEO DEG FDR 5% via subseries GSE107334)	HepG2 cells treated with 10 nM insulin versus no insulin for 4 h	4137	3424	1002 ($2.74\times {10}^{-33}$)
Insulin	Insulin Receptor Associates Promoters GSE107336 (Hancock et al., 2019) (Enrichr geneset)	HepG2 cells treated with 10 nM insulin versus no insulin for 4 h	457	3424	103 (0.00471)
Estrogen	Estrogen human MCF-7 cells GSE11324 (Carroll et al., 2006) (Enrichr genesets ligand: 89,90,91)	MCF7 cells exposed to 100 nM estrogen for 3, 6 or 12 h versus 0 h	953	9276	607 ($1.40\times {10}^{-23}$)
Estrogen	Estradiol human estrogen receptor (ER)-positive MCF7 breast cancer cells GDS3217 (Lin et al., 2007) (Enrichr genesets ligand: 39,40,41)	MCF7 cells exposed to 10 nM estradiol versus vehicle-only at 12, 24 and 48 h	869	9276	487 ($7.54\times {10}^{-7}$)

Note: GEO study identifiers are prefixed by ‘GSE’ or ‘GDS’. For insulin, besides the ref. geneset from Enrichr, the DEG table in GEO was used to derive another ref. geneset for the same experimental study at 5% false discovery rate (FDR). For estrogen, the DEGs in all timepoints of a study are combined to get a single ref. geneset per study.