The rise of high-throughput methods has allowed for screens with millions of single compounds and thousands of (binary) compound combinations. Investigating compounds in combination is cost and labor intensive. This has motivated recent investigations to predict synergistic combinations from information of the compound targets, the known pathways and the gene expression response to single compounds. Even though these predictive models are reasonably successful (AUCs around 0.9), there is room for improvement. Importantly, these models are often evaluated with data sets collected for other purposes. In particular, there exists currently no large-scale data set of molecular events underlying synergy. Such data would allow for an improvement of synergy prediction from molecular data in terms of model validation and development. The L1000 platform has proven to be a powerful tool for gene expression based hypothesis generation and small molecule screening. The assay has been used extensively to investigate the transcriptional signatures of small molecule perturbations, but there is not yet a well-validated experimental workflow for studying compounds in combination. We combined a cell-viability readout with an L1000-based gene expression readout on a diverse set of compounds.

We implemented the code to fit dose-response curves to L1000 data and phenotype readouts such as from a Cell-Titer Glow assay or Incucyte. The code contains many visualization tools to visualize the plate-layout of 384 well plates, fit dose-response curves of the drc or GRdrc of L1000 or phenotype data, apply outlier detection based on GAMs or dr4pl method, and other functions useful for the analysis of cell viability data and gene expression.

This code was generated for the analyses conducted in "Lederer, S., Lyons, N., Dijkstra, (unpublished). 'Investigating Compound Interactions - from Gene Expression to Phenotype', unpubished work"