Fig 1.
(A) A one-drug-multi-target-multi-indication strategy to screen drugs that can both enhance therapeutic effect and mitigate side effect. (B) Schema of 3D-REMAP, a multi-target screening platform that integrates structural genomics and chemical genomics data and combines tools from bioinformatics, chemoinformatics, protein-ligand docking, and machine learning. R and Q denote observed protein-chemical interactions in chemical genomics databases, and predicted protein-chemical interactions from ligand binding site similarity coupled with protein-ligand docking, respectively. These two matrices are the input for the machine learning algorithm weighted imputed neighborhood-regularized One-Class Collaborative Filtering (winOCCF) to predict genome-wide drug-target interactions. See Method section for details. DTI: drug-target interaction.
Table 1.
Top 20 ranked putative off-targets of PDE3B.
Docking score that is less than -7.5 is highlighted in bold.
Fig 2.
The distribution of kinase off-targets of levosimendan in the human kinome.
The off-targets are marked by red circles. The diameter of the circles approximately corresponds to the binding strength. Illustration reproduced courtesy of Cell Signaling Technology, Inc. (www.cellsignal.com/).
Fig 3.
(A) Predicted binding poses of levosimendan (blue stick) and co-crystallized ADP (yellow stick) on RIOK1 (ribbon model). (B) Interaction pattern of levosimendan with RIOK1.
Table 2.
Comparison of the performance of 3D-REMAP with other methods when predicting that kinase off-targets of levosimendan, which are ranked at the top 2.5%.
Fig 4.
The drug dose response curve of lymphoma SU-DHL-8 cell line under the treatment of levosimendan.
Table 3.
Overrepresented GO biological process terms responsible for the anti-cancer sensitivity of levosimendan.
Fig 5.
Percentage of cases that are ranked within top 100 in the predictive model over all cases in the TCGA project.
The most statistically significant overrepresented cancer type is B-cell lymphoma (TCGA-DLBC).