Limits...
Recovering drug-induced apoptosis subnetwork from Connectivity Map data.

Yu J, Putcha P, Silva JM - Biomed Res Int (2015)

Bottom Line: We then focused on 13 apoptotic genes that showed significant differential expression across all drug-perturbed samples to reconstruct the apoptosis network.In our predicted subnetwork, 9 out of 15 high-confidence interactions were validated in the literature, and our inferred network captured two major cell death pathways by identifying BCL2L11 and PMAIP1 as key interacting players for the intrinsic apoptosis pathway and TAXBP1 and TNFAIP3 for the extrinsic apoptosis pathway.Our inferred apoptosis network also suggested the role of BCL2L11 and TNFAIP3 as "gateway" genes in the drug-induced intrinsic and extrinsic apoptosis pathways.

View Article: PubMed Central - PubMed

Affiliation: Department of Precision Medicine, Oncology Research Unit, Pfizer Inc., Pearl River, NY 10965, USA.

ABSTRACT
The Connectivity Map (CMAP) project profiled human cancer cell lines exposed to a library of anticancer compounds with the goal of connecting cancer with underlying genes and potential treatments. Since the therapeutic goal of most anticancer drugs is to induce tumor-selective apoptosis, it is critical to understand the specific cell death pathways triggered by drugs. This can help to better understand the mechanism of how cancer cells respond to chemical stimulations and improve the treatment of human tumors. In this study, using CMAP microarray data from breast cancer cell line MCF7, we applied a Gaussian Bayesian network modeling approach and identified apoptosis as a major drug-induced cellular-pathway. We then focused on 13 apoptotic genes that showed significant differential expression across all drug-perturbed samples to reconstruct the apoptosis network. In our predicted subnetwork, 9 out of 15 high-confidence interactions were validated in the literature, and our inferred network captured two major cell death pathways by identifying BCL2L11 and PMAIP1 as key interacting players for the intrinsic apoptosis pathway and TAXBP1 and TNFAIP3 for the extrinsic apoptosis pathway. Our inferred apoptosis network also suggested the role of BCL2L11 and TNFAIP3 as "gateway" genes in the drug-induced intrinsic and extrinsic apoptosis pathways.

Show MeSH

Related in: MedlinePlus

(a) Histogram and (b) box plot of scores for best-learned graphical model in each bootstrapped sampling.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4389823&req=5

fig5: (a) Histogram and (b) box plot of scores for best-learned graphical model in each bootstrapped sampling.

Mentions: With the methods outlined above, we obtained a Bayesian network structure that best described the observed data. However, it is possible that the model may be overfitted, which means that a small change to the dataset could make the network structure change dramatically. A way to solve this issue is to apply a resampling method or simulating the dataset. The method would learn the best graphical model for each sampled dataset and generate a consensus network from the average of the sample models. This method is also known as model averaging. The simulation method we used to do model averaging was Efron's bootstrapping method [29, 30]. To increase robustness, the method only considered predicted network structures with a score within 95% of the confidence interval. The distribution of network scores is shown in Figure 5. In generating the final combined consensus network, edges were selected based on a confidence threshold of 75%.


Recovering drug-induced apoptosis subnetwork from Connectivity Map data.

Yu J, Putcha P, Silva JM - Biomed Res Int (2015)

(a) Histogram and (b) box plot of scores for best-learned graphical model in each bootstrapped sampling.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4389823&req=5

fig5: (a) Histogram and (b) box plot of scores for best-learned graphical model in each bootstrapped sampling.
Mentions: With the methods outlined above, we obtained a Bayesian network structure that best described the observed data. However, it is possible that the model may be overfitted, which means that a small change to the dataset could make the network structure change dramatically. A way to solve this issue is to apply a resampling method or simulating the dataset. The method would learn the best graphical model for each sampled dataset and generate a consensus network from the average of the sample models. This method is also known as model averaging. The simulation method we used to do model averaging was Efron's bootstrapping method [29, 30]. To increase robustness, the method only considered predicted network structures with a score within 95% of the confidence interval. The distribution of network scores is shown in Figure 5. In generating the final combined consensus network, edges were selected based on a confidence threshold of 75%.

Bottom Line: We then focused on 13 apoptotic genes that showed significant differential expression across all drug-perturbed samples to reconstruct the apoptosis network.In our predicted subnetwork, 9 out of 15 high-confidence interactions were validated in the literature, and our inferred network captured two major cell death pathways by identifying BCL2L11 and PMAIP1 as key interacting players for the intrinsic apoptosis pathway and TAXBP1 and TNFAIP3 for the extrinsic apoptosis pathway.Our inferred apoptosis network also suggested the role of BCL2L11 and TNFAIP3 as "gateway" genes in the drug-induced intrinsic and extrinsic apoptosis pathways.

View Article: PubMed Central - PubMed

Affiliation: Department of Precision Medicine, Oncology Research Unit, Pfizer Inc., Pearl River, NY 10965, USA.

ABSTRACT
The Connectivity Map (CMAP) project profiled human cancer cell lines exposed to a library of anticancer compounds with the goal of connecting cancer with underlying genes and potential treatments. Since the therapeutic goal of most anticancer drugs is to induce tumor-selective apoptosis, it is critical to understand the specific cell death pathways triggered by drugs. This can help to better understand the mechanism of how cancer cells respond to chemical stimulations and improve the treatment of human tumors. In this study, using CMAP microarray data from breast cancer cell line MCF7, we applied a Gaussian Bayesian network modeling approach and identified apoptosis as a major drug-induced cellular-pathway. We then focused on 13 apoptotic genes that showed significant differential expression across all drug-perturbed samples to reconstruct the apoptosis network. In our predicted subnetwork, 9 out of 15 high-confidence interactions were validated in the literature, and our inferred network captured two major cell death pathways by identifying BCL2L11 and PMAIP1 as key interacting players for the intrinsic apoptosis pathway and TAXBP1 and TNFAIP3 for the extrinsic apoptosis pathway. Our inferred apoptosis network also suggested the role of BCL2L11 and TNFAIP3 as "gateway" genes in the drug-induced intrinsic and extrinsic apoptosis pathways.

Show MeSH
Related in: MedlinePlus