DePick is a novel target de-convolution tool to determine targets specifically linked to a phenotype mearsured in a high-throughput screening assay, such as lipid transferring and Wnt signaling inhibition. This method first identifies the compounds with specific activity on a screening assay, and then determines their protein targets that are statistically enriched in this set of compounds. The target determination approach is based on HitPick (Liu et al, 2013), a recently developed in silico target prediction method, coupled to the identification of the statistically enriched targets of the set of specific hits.
DePick target de-convolution scheme:
Given the pre-defined “experiment” and “control” screening assays, DePick will firstly apply the modified B-Score method (Liu & Campillos, 2014) to identify hits for each assay, subsequently separate the tested compounds into two sets of compounds containing “Specific hits” and “Inactive compounds” (Figure 1). The “Specific hits” set is composed of the compounds that modulate specifically the phenotypes of interest, that is, the compounds active in the “experiment” assay and inactive on the “control” assay. The compounds inactive in the experiment assay form the “Inactive compounds” set. We subsequently applied the hypergeometric test to detect predicted target(s) that are over-represented in the “Specific hits” set when compared to “Inactive compounds” set. We associated to the phenotype those protein targets with a resulting p-value lower than 0.05 after false discovery rate (FDR) multiple testing correction.
DePick webserver also allow users to submit the compounds data which contains “Specific hits” and “Inactive compounds” to detect the predicted targets for the assays.
Fig.1 Steps of DePick method for the Identification of drug targets associated to phenotypic screens.
The format of the screening data in DePick is the same used by ChemBank. As an example, we provide the following chemical screening from ChemBank. The uploaded screening data from user should be tab delimited ".txt" file.
If there is only one assay, DePick will compare the hits and inactive compounds of this assay (download example). If there are multiple “experiment” and “control” assays, DePick will join the hits or inactive compounds, respectively, from all the corresponding “experiment” and “control” assays (download example).
Compounds data are “Specific hits” and “Inactive compounds” sets - two lists compounds with their SMILES strings separated by tab.
The output of the DePick web server is list of target predictions which are significantly over-represented in the high-throughput chemical phenotypic screen. Below are the 5 targets that are enriched in lipid transferring assay (Figure 2). For more detailed information, we also provide users the specific hits set of the assay and the whole list of the targets with the q-values.
Fig.2 Analysis of “SR-BI lipid transfer” screen. Relationships between significantly predicted protein targets (purple and orange rectangles) and the lipid transfer mediated by SR-B1 transporter phenotype assay. This interaction network has been created using the CIDeR database (Lechner et al, 2012) where all the human gene names are provided by EntrezGene. Beige and green rectangles indicate background proteins and chemicals related to the assay. An expanded view of this network can be visualized following this link (http://mips.helmholtz-muenchen.de/cider/SRBI). For reasons of clarity and comprehensibility the graph contains only a part of the information that is available in the literature.
DePick processing time depends on the size of the assay and tested compounds data. For bioassays size lower than 100,000 compounds, the web server returns the results in ~ 20 minutes.
However, as calculations are carried out on a shared cluster environment, actual processing time depends on the cluster workload.
Liu X, Vogt I, Haque T & Campillos M (2013) HitPick: a web server for hit identification and target prediction of chemical screenings. Bioinformatics 29: 1910–1912.
Liu X & Campillos M (2014) Unveiling new biological relationships using shared hits of chemical screening assay pairs. Bioinformatics 30: i579–i586
Lechner M, Höhn V, Brauner B, Dunger I, Fobo G, Frishman G, Montrone C, Kastenmüller G, Waegele B & Ruepp A (2012) CIDeR: multifactorial interaction networks in human diseases. Genome Biol. 13: R62