Limits...
NEMix: single-cell nested effects models for probabilistic pathway stimulation.

Siebourg-Polster J, Mudrak D, Emmenlauer M, Rämö P, Dehio C, Greber U, Fröhlich H, Beerenwinkel N - PLoS Comput. Biol. (2015)

Bottom Line: Nested effects models have been used successfully for learning subcellular networks from high-dimensional perturbation effects that result from RNA interference (RNAi) experiments.As a consequence of this cellular heterogeneity, knock-downs result in variable effects among cells and lead to weak average phenotypes on the cell population level.Using a subset of genes with known interactions, we show that the inferred NEMix network has high accuracy and outperforms the classical nested effects model without hidden pathway activity.

View Article: PubMed Central - PubMed

Affiliation: Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland; SIB Swiss Institute of Bioinformatics, Basel, Switzerland.

ABSTRACT
Nested effects models have been used successfully for learning subcellular networks from high-dimensional perturbation effects that result from RNA interference (RNAi) experiments. Here, we further develop the basic nested effects model using high-content single-cell imaging data from RNAi screens of cultured cells infected with human rhinovirus. RNAi screens with single-cell readouts are becoming increasingly common, and they often reveal high cell-to-cell variation. As a consequence of this cellular heterogeneity, knock-downs result in variable effects among cells and lead to weak average phenotypes on the cell population level. To address this confounding factor in network inference, we explicitly model the stimulation status of a signaling pathway in individual cells. We extend the framework of nested effects models to probabilistic combinatorial knock-downs and propose NEMix, a nested effects mixture model that accounts for unobserved pathway activation. We analyzed the identifiability of NEMix and developed a parameter inference scheme based on the Expectation Maximization algorithm. In an extensive simulation study, we show that NEMix improves learning of pathway structures over classical NEMs significantly in the presence of hidden pathway stimulation. We applied our model to single-cell imaging data from RNAi screens monitoring human rhinovirus infection, where limited infection efficiency of the assay results in uncertain pathway stimulation. Using a subset of genes with known interactions, we show that the inferred NEMix network has high accuracy and outperforms the classical nested effects model without hidden pathway activity. NEMix is implemented as part of the R/Bioconductor package 'nem' and available at www.cbg.ethz.ch/software/NEMix.

No MeSH data available.


Related in: MedlinePlus

Performance comparison of the simulations.(A) Simulation results are summarized based on the accuracy of recovered edges for the compared methods. The methods are random, random edge sampling with rate ; NEM, the normal NEM inference; sc-NEM, the cell level NEM and NEMix, using the NEMix inference with the hidden pathway state. All methods were run on 50 simulated data sets from 30 sample networks, repeated for different knock-down probabilities of the pathway state p0. (B) For the NEMix model, the distributions of inferred p0 values are compared to the true p0.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4400057&req=5

pcbi.1004078.g002: Performance comparison of the simulations.(A) Simulation results are summarized based on the accuracy of recovered edges for the compared methods. The methods are random, random edge sampling with rate ; NEM, the normal NEM inference; sc-NEM, the cell level NEM and NEMix, using the NEMix inference with the hidden pathway state. All methods were run on 50 simulated data sets from 30 sample networks, repeated for different knock-down probabilities of the pathway state p0. (B) For the NEMix model, the distributions of inferred p0 values are compared to the true p0.

Mentions: Fig. 2A summarizes the overall performance for all methods and the different fractions of pathway signal perturbation p0. We display accuracy of the edge recovery, for varying p0. We also calculated the area under the ROC curve (AUC) based on the edge frequencies of the 50 replicate data sets, which yielded similar results in terms of accuracy (see supplementary S2 Fig). As expected, all methods performed equally well when there is no signal disruption (p0 = 0). However, when p0 is moderate to high, NEMix performs significantly better than the other methods. If the triggering signal is always turned off, performance of all methods drops drastically. Intuitively, this is because in such a special case, all features downstream of Z always show an effect and hence they cannot be used for structure learning. For example, if, in Fig. 1B, Z is inactive for each cell, we could not infer the structure among S2 and S3. In reality though, permanent shut down of the pathway is very unlikely. For the infection screens p0 = 1 would mean that no cell is ever infected. Pathway activity estimates are also of overall high accuracy (Fig. 2B). Although simulation results demonstrate that the performance of learning Z and θ varies, depending on the network structure, the average performance is very good (S3 Fig, S4 Fig, S5 Fig).


NEMix: single-cell nested effects models for probabilistic pathway stimulation.

Siebourg-Polster J, Mudrak D, Emmenlauer M, Rämö P, Dehio C, Greber U, Fröhlich H, Beerenwinkel N - PLoS Comput. Biol. (2015)

Performance comparison of the simulations.(A) Simulation results are summarized based on the accuracy of recovered edges for the compared methods. The methods are random, random edge sampling with rate ; NEM, the normal NEM inference; sc-NEM, the cell level NEM and NEMix, using the NEMix inference with the hidden pathway state. All methods were run on 50 simulated data sets from 30 sample networks, repeated for different knock-down probabilities of the pathway state p0. (B) For the NEMix model, the distributions of inferred p0 values are compared to the true p0.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4400057&req=5

pcbi.1004078.g002: Performance comparison of the simulations.(A) Simulation results are summarized based on the accuracy of recovered edges for the compared methods. The methods are random, random edge sampling with rate ; NEM, the normal NEM inference; sc-NEM, the cell level NEM and NEMix, using the NEMix inference with the hidden pathway state. All methods were run on 50 simulated data sets from 30 sample networks, repeated for different knock-down probabilities of the pathway state p0. (B) For the NEMix model, the distributions of inferred p0 values are compared to the true p0.
Mentions: Fig. 2A summarizes the overall performance for all methods and the different fractions of pathway signal perturbation p0. We display accuracy of the edge recovery, for varying p0. We also calculated the area under the ROC curve (AUC) based on the edge frequencies of the 50 replicate data sets, which yielded similar results in terms of accuracy (see supplementary S2 Fig). As expected, all methods performed equally well when there is no signal disruption (p0 = 0). However, when p0 is moderate to high, NEMix performs significantly better than the other methods. If the triggering signal is always turned off, performance of all methods drops drastically. Intuitively, this is because in such a special case, all features downstream of Z always show an effect and hence they cannot be used for structure learning. For example, if, in Fig. 1B, Z is inactive for each cell, we could not infer the structure among S2 and S3. In reality though, permanent shut down of the pathway is very unlikely. For the infection screens p0 = 1 would mean that no cell is ever infected. Pathway activity estimates are also of overall high accuracy (Fig. 2B). Although simulation results demonstrate that the performance of learning Z and θ varies, depending on the network structure, the average performance is very good (S3 Fig, S4 Fig, S5 Fig).

Bottom Line: Nested effects models have been used successfully for learning subcellular networks from high-dimensional perturbation effects that result from RNA interference (RNAi) experiments.As a consequence of this cellular heterogeneity, knock-downs result in variable effects among cells and lead to weak average phenotypes on the cell population level.Using a subset of genes with known interactions, we show that the inferred NEMix network has high accuracy and outperforms the classical nested effects model without hidden pathway activity.

View Article: PubMed Central - PubMed

Affiliation: Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland; SIB Swiss Institute of Bioinformatics, Basel, Switzerland.

ABSTRACT
Nested effects models have been used successfully for learning subcellular networks from high-dimensional perturbation effects that result from RNA interference (RNAi) experiments. Here, we further develop the basic nested effects model using high-content single-cell imaging data from RNAi screens of cultured cells infected with human rhinovirus. RNAi screens with single-cell readouts are becoming increasingly common, and they often reveal high cell-to-cell variation. As a consequence of this cellular heterogeneity, knock-downs result in variable effects among cells and lead to weak average phenotypes on the cell population level. To address this confounding factor in network inference, we explicitly model the stimulation status of a signaling pathway in individual cells. We extend the framework of nested effects models to probabilistic combinatorial knock-downs and propose NEMix, a nested effects mixture model that accounts for unobserved pathway activation. We analyzed the identifiability of NEMix and developed a parameter inference scheme based on the Expectation Maximization algorithm. In an extensive simulation study, we show that NEMix improves learning of pathway structures over classical NEMs significantly in the presence of hidden pathway stimulation. We applied our model to single-cell imaging data from RNAi screens monitoring human rhinovirus infection, where limited infection efficiency of the assay results in uncertain pathway stimulation. Using a subset of genes with known interactions, we show that the inferred NEMix network has high accuracy and outperforms the classical nested effects model without hidden pathway activity. NEMix is implemented as part of the R/Bioconductor package 'nem' and available at www.cbg.ethz.ch/software/NEMix.

No MeSH data available.


Related in: MedlinePlus