Algorithmic pipeline.

<p>(<b>A</b>) Mapping drugs onto PPI network through drug-target associations. (<b>B</b>) Weighting PPI network by gene expression profile across human tissues. (<b>C</b>) Defining target-centered system, including known drug-targets and their first-step neighbor protein on PPI network. (<b>D</b>) Scoring systematic connectivity based on drugs' target-centered systems (S-score). (<b>E</b>) Scoring phenotypic similarity based on drug clinical side effects (P-score). (<b>F</b>) Cross-validating between S-score and P-score. (<b>G</b>) Integrating by a Bayesian probabilistic model. (<b>H</b>) Validating the prediction performance using GSP DDIs from two independent databases, DrugBank and STITCH. (<b>I</b>) Comparing with other methods. (<b>J</b>) Explaining the molecular mechanisms of the polypharmacy side effects recorded in TWOSIDES.