Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2021 Apr 22:12:608042.
doi: 10.3389/fgene.2021.608042. eCollection 2021.

NEM-Tar: A Probabilistic Graphical Model for Cancer Regulatory Network Inference and Prioritization of Potential Therapeutic Targets From Multi-Omics Data

Yuchen Zhang  1 , Lina Zhu  1 , Xin Wang  1   2
Affiliations

NEM-Tar: A Probabilistic Graphical Model for Cancer Regulatory Network Inference and Prioritization of Potential Therapeutic Targets From Multi-Omics Data

Yuchen Zhang et al. Front Genet. .

Abstract

Targeted therapy has been widely adopted as an effective treatment strategy to battle against cancer. However, cancers are not single disease entities, but comprising multiple molecularly distinct subtypes, and the heterogeneity nature prevents precise selection of patients for optimized therapy. Dissecting cancer subtype-specific signaling pathways is crucial to pinpointing dysregulated genes for the prioritization of novel therapeutic targets. Nested effects models (NEMs) are a group of graphical models that encode subset relations between observed downstream effects under perturbations to upstream signaling genes, providing a prototype for mapping the inner workings of the cell. In this study, we developed NEM-Tar, which extends the original NEMs to predict drug targets by incorporating causal information of (epi)genetic aberrations for signaling pathway inference. An information theory-based score, weighted information gain (WIG), was proposed to assess the impact of signaling genes on a specific downstream biological process of interest. Subsequently, we conducted simulation studies to compare three inference methods and found that the greedy hill-climbing algorithm demonstrated the highest accuracy and robustness to noise. Furthermore, two case studies were conducted using multi-omics data for colorectal cancer (CRC) and gastric cancer (GC) in the TCGA database. Using NEM-Tar, we inferred signaling networks driving the poor-prognosis subtypes of CRC and GC, respectively. Our model prioritized not only potential individual drug targets such as HER2, for which FDA-approved inhibitors are available but also the combinations of multiple targets potentially useful for the design of combination therapies.

Keywords: cancer; combination therapy; drug targets; molecular subtype; nested effects model; regulatory network.

PubMed Disclaimer

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Figures

FIGURE 1
FIGURE 1
The workflow of NEM-Tar for cancer regulatory network inference and potential drug targets prioritization. Observations of the states of S-genes and E-genes could be obtained after the preprocessing of multi-omics data. The signaling network regulating a specific cancer subtype will subsequently be inferred. Finally, based on quantification of the causal impact and specificity to downstream genes using WIG, potential drug targets could be prioritized for single and double perturbations.
FIGURE 2
FIGURE 2
Illustration of the nested effects model and NEM-Tar for real cancer samples. (A) The S-genes are modeled as hidden variables, and their signaling interaction graph G (solid arrows) is the target to infer. In experiments with perturbations to individual S-genes, differential expression of downstream genes could be observed and considered as effect reporter genes (E-genes). Assuming that each E-gene is directly regulated by at most one S-gene in G, the maximum a posteriori attachment Θ (dashed arrows) of effect genes to S-genes could be computed. The goal is to search for the signaling graph G, which yields the most likely probabilistic nested effects. (B) For an extra observational dimension (the real patients), the necessary adjustment should be conducted on the design and inference strategies of classic NEM. However, the information that needs to be inferred is also the hidden interaction between S-genes and the attachment relationship of E-genes to S-genes.
FIGURE 3
FIGURE 3
Illustration for the definition of Weighted Information Gain (WIG) using a toy example. (A) A toy network containing four S-genes with their corresponding E-gene attachment. Note that the hierarchies of S1 and S2 genes cannot be distinguished. (B) Posterior effect positions obtained after network inference; (C) Uniformly distributed effect positions before inference. Suppose that the attached E-genes to a S-gene are all signature genes related to a pathway of interest (e.g., EMT), it could be easily calculated that S4 has the highest causal impact on the particular downstream pathway, and S1 and S2 have the same impact. As an example, we illustrated the calculation of WIG(S3).
FIGURE 4
FIGURE 4
A comparison of the performance of three representative network inference strategies. (A–C) The performance of NEM-Tar based on (A) MCMC sampling, (B) triple relations, and (C) greedy hill-climbing, respectively, on simulated data for varying numbers of S-genes. For each method, we generated 200 random signaling networks and inferred their structures using NEM-Tar from the simulated E-gene data. (D) The performance of NEM-Tar based on greedy hill-climbing testing its robustness to simulated data with different levels of noise.
FIGURE 5
FIGURE 5
The case studies of NEM-Tar on gastric cancer and colorectal cancer. (A) Reconstructed signaling network for the EMT subtype of gastric cancer. (B) Reconstructed signaling network for the CMS4 subtype of colorectal cancer.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Anastas J. N., Moon R. T. (2013). WNT signalling pathways as therapeutic targets in cancer. Nat. Rev. Cancer 13 11–26. 10.1038/nrc3419 - DOI - PubMed
    1. Anchang B., Sadeh M. J., Jacob J., Tresch A., Vlad M. O., Oefner P. J., et al. (2009). Modeling the temporal interplay of molecular signaling and gene expression by using dynamic nested effects models. Proc. Natl. Acad. Sci. U. S. A. 106 6447–6452. 10.1073/pnas.0809822106 - DOI - PMC - PubMed
    1. Andrieu C., de Freitas N., Doucet A., Jordan M. I. (2003). An Introduction to MCMC for Machine Learning. Mach. Learn. 50 5–43. 10.1023/a:1020281327116 - DOI
    1. Bang Y.-J., Van Cutsem E., Feyereislova A., Chung H. C., Shen L., Sawaki A., et al. (2010). Trastuzumab in combination with chemotherapy versus chemotherapy alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet 376 687–697. 10.1016/S0140-6736(10)61121-X - DOI - PubMed
    1. Bashashati A., Haffari G., Ding J., Ha G., Lui K., Rosner J., et al. (2012). DriverNet: uncovering the impact of somatic driver mutations on transcriptional networks in cancer. Genome Biol. 13:R124. 10.1186/gb-2012-13-12-r124 - DOI - PMC - PubMed