I'm primarily interested in creating structure-activity relationships (SAR) for compounds, with differential gene expression as the target (e.g. given a compound, predict the differential expression (likely just a classification of up/down-regulated) of all human genes . There are a number of papers that have have made interesting approaches (https://academic.oup.com/bioinformatics/article/32/12/1832/1743989, and http://pubs.acs.org/doi/abs/10.1021/acs.jcim.6b00260), and I'd like to make a similar approach. Optimally, the dataset would have similar experimental conditions for a large number of genes and compounds tested. I've found the GEO Omnibus (https://www.ncbi.nlm.nih.gov/gds), but its unclear if this is the collection of datasets that I care about.

Edit: to be clear, I'm hoping for datasets that exist for each prominent cell lines (e.g. NCI-60, or something similar).