Academic Commons

Theses Doctoral

Interrogating Drug Mechanism of Action Using Network Dysregulation Analysis

Woo, Junghoon

Accurate identification of small-molecule compound substrates and effectors, within specific tissues, represents a highly relevant yet equally elusive objective. Accomplishing this goal would have major implications on the assessment of compound efficacy and potential toxicity with significant impact on drug discovery and development. Computationally, there are no methods to elucidate a compound mechanisms of action (MoA) in cell-context-specific and genome-wide fashions. Experimental approaches are equally limited in that they are effective in identifying only specific drug substrate classes (e.g., high-affinity substrates of kinase inhibitors) rather than the full repertoire of proteins that effect compound activity in a specific tissue, including those that may cause undesired toxicity. They are costly, laborious, and the relevant mechanistic assays can only be performed in vitro.
Here I introduce DeMAND, a novel algorithm for the regulatory network-based elucidation of compound Mechanisms of Action. The algorithm interrogates a context-specific regulatory network using at least six gene-expression profiles representative of in vitro or in vivo compound perturbation to identify compound dysregulated sub-networks as well as substrates and effector proteins. In experimental tests, the algorithm correctly identified proteins in the established MoA of over 90% of the tested compounds, including protein such as SIK1, a private effector of doxorubicin responsible for its cardiac toxicity, which is however not affected by less toxic topoisomerase inhibitors, such as camptothecin. Using gene expression profiles following perturbation of diffuse large B cell lymphoma cells with 14 and 92 compounds, respectively, at different concentrations and time points, I identified and validated several novel effector proteins. These include RPS3A (ribosomal protein S3A), VHL (von Hippel-Lindau tumor suppressor, E3 ubiquitin protein ligase), and CCNB1 (cyclin B1) as effectors of the mitotic spindle inhibitor vincristine, all of which significantly affected microtubule architecture and/or modulated vincristine activity when silenced, as well as JAK2 (Janus kinase 2) as a novel effector/modulator of mitomycin C, which desensitizes cells to mitomycin C treatment when silenced.
Finally, I used DeMAND to evaluate compound similarity by comparing the proteins in their MoA. I tested the similarity of altretamine, a compound with currently unknown substrates, and sulfasalazine, which were predicted to have similar MoA and in particular to be inhibitors of the GPX4 (glutathione peroxidase 4) protein. Experimental validation confirmed this prediction as well as increase in lipid reactive oxygen species (ROS) levels, a recently established downstream effector of sulfasalazine.
Critically, DeMAND suggests that regulatory networks reverse engineered de novo form large molecular profile datasets can provide novel mechanistic insight into drug activity, thus providing a significant novel contribution to our search for highly specific and non-toxic small-molecule inhibitors.


  • thumnail for WOO_columbia_0054D_12545.pdf WOO_columbia_0054D_12545.pdf binary/octet-stream 4.93 MB Download File

More About This Work

Academic Units
Biomedical Informatics
Thesis Advisors
Califano, Andrea
Ph.D., Columbia University
Published Here
March 6, 2015