Path4Drug: Data Science Workflow for Identification of Tissue-Specific Biological Pathways Modulated by Toxic Drugs

Abstract

The early prediction of drug adverse effects is of great interest to pharmaceutical research, as toxicity is one of the leading reasons for drug attrition. Understanding the cell signaling and regulatory pathways affected by a drug candidate is crucial to the study of drug toxicity. In this study, we present a computational technique that employs the propagation of drug-protein interactions to connect compounds to biological pathways. Target profiles for drugs were built by retrieving drug target proteins from public repositories such as ChEMBL, DrugBank, IUPHAR, PharmGKB, and TTD. Subsequent enrichment test of the protein pool using Reactome revealed potential pathways affected by the drugs. Furthermore, an optional tissue filter utilizing the Human Protein Atlas was applied to identify tissue-specific pathways. The analysis pipeline was implemented in an open-source KNIME workflow called Path4Drug to allow automated data retrieval and reconstruction for any given drug present in ChEMBL. The pipeline was applied to withdrawn drugs and cardio- and hepatotoxic drugs with black box warnings to identify biochemical pathways they affect and to find pathways that can be potentially connected to the toxic events. To complement this approach, drugs used in cardiac therapy without any record of toxicity were also analyzed. The results provide already known associations as well as a large amount of additional potential connections. Consequently, our approach can link drugs to biological pathways by leveraging big data available in public resources. The developed tool is openly available and modifiable to support other systems biology analyses.

Unravelling Mechanisms of Doxorubicin-Induced Toxicity in 3D Human Intestinal Organoids

“…Protein type targets of DOX were collected from open repositories such as ChEMBL [25], DrugBank [26], TTD [27], IUPHAR [28], PharmGKB [24] using a pre-constructed KNIME data science workflow [23]. The criteria for a target protein were individually adjusted to the guidelines of the corresponding database.…”

“…Cytotoxicity measurements were evaluated and checked if they were in line with the transcriptomic responses. In addition, proteomics data were assessed through computational simulations [23,24] based on known DOX protein targets in the gut available in online repositories [25][26][27][28]. These data were useful to assess if target proteins were reflected on the transcriptomic data generated on the organoids, as well as to gain a broader insight into the drug mechanisms of toxicity.…”

KNIME workflow for retrieving causal drug and protein interactions, building networks, and performing topological enrichment analysis demonstrated by a DILI case study

“…In the first section of the workflow, causal target protein profiles can be built using three different databases. In a previous publication, a detailed description of retrieving target profiles for compounds was described [ 14 ]. However, as the previous workflow does not distinguish between positive and negative effects of the compounds, in this workflow this type of information is emphasized: Targets were defined as human-type single proteins, which are annotated as target proteins of the compound of interest in one of the utilized databases or the compound was annotated as active on the protein in biological assays or the compound has an activity value to the target in a pre-defined active range.…”