Limits...
Context-Specific Metabolic Model Extraction Based on Regularized Least Squares Optimization.

Robaina Estévez S, Nikoloski Z - PLoS ONE (2015)

Bottom Line: Here we present RegrEx, a fully automated approach that relies solely on context-specific data and ℓ1-norm regularization to extract a context-specific model and to provide a flux distribution that maximizes its correlation to data.Moreover, the publically available implementation of RegrEx was used to extract 11 context-specific human models using publicly available RNAseq expression profiles, Recon1 and also Recon2, the most recent human metabolic model.Therefore, our study sets the ground for applications of other regularization techniques in large-scale metabolic modeling.

View Article: PubMed Central - PubMed

Affiliation: Systems Biology and Mathematical Modeling Group, Max Planck Institute of Molecular Plant Physiology (MPIMP), Potsdam-Golm, Germany.

ABSTRACT
Genome-scale metabolic models have proven highly valuable in investigating cell physiology. Recent advances include the development of methods to extract context-specific models capable of describing metabolism under more specific scenarios (e.g., cell types). Yet, none of the existing computational approaches allows for a fully automated model extraction and determination of a flux distribution independent of user-defined parameters. Here we present RegrEx, a fully automated approach that relies solely on context-specific data and ℓ1-norm regularization to extract a context-specific model and to provide a flux distribution that maximizes its correlation to data. Moreover, the publically available implementation of RegrEx was used to extract 11 context-specific human models using publicly available RNAseq expression profiles, Recon1 and also Recon2, the most recent human metabolic model. The comparison of the performance of RegrEx and its contending alternatives demonstrates that the proposed method extracts models for which both the structure, i.e., reactions included, and the flux distributions are in concordance with the employed data. These findings are supported by validation and comparison of method performance on additional data not used in context-specific model extraction. Therefore, our study sets the ground for applications of other regularization techniques in large-scale metabolic modeling.

No MeSH data available.


Illustration of selected Recon 1 subsystems displayed in a Pie chart form, depicting the distribution of mean flux capacity (MFC) values across contexts.Panels A-B correspond to the two extreme metabolic subsystems, in terms of CV, in Central metabolism. The citric acid cycle (A) shows the lowest CV value (both in Central metabolism and within the entirety of Recon 1 subsystems). The pentose phosphate pathway (B) shows the greatest CV value in Central metabolism. (C-E) the distribution of MFC values is shown for fatty acid metabolism (C) which is predominantly represented in liver, adipose tissue and skeletal muscle, fatty acid oxidation (D) and fatty acid activation (E) both subsystems predominant in adipose tissue and skeletal muscle. (F) The MFC distribution across contexts is depicted for gluthatione metabolism. Kidney is the context where this subsystem gets a highest MFC value, constituting a 23% of the total MFC value across contexts. See main text for details. In all cases, the first number preceding the name of the subsystem corresponds to its position in the ranking generated by the CV values, which are shown in round brackets here. Context names are displayed in the color bar legend.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4497637&req=5

pone.0131875.g003: Illustration of selected Recon 1 subsystems displayed in a Pie chart form, depicting the distribution of mean flux capacity (MFC) values across contexts.Panels A-B correspond to the two extreme metabolic subsystems, in terms of CV, in Central metabolism. The citric acid cycle (A) shows the lowest CV value (both in Central metabolism and within the entirety of Recon 1 subsystems). The pentose phosphate pathway (B) shows the greatest CV value in Central metabolism. (C-E) the distribution of MFC values is shown for fatty acid metabolism (C) which is predominantly represented in liver, adipose tissue and skeletal muscle, fatty acid oxidation (D) and fatty acid activation (E) both subsystems predominant in adipose tissue and skeletal muscle. (F) The MFC distribution across contexts is depicted for gluthatione metabolism. Kidney is the context where this subsystem gets a highest MFC value, constituting a 23% of the total MFC value across contexts. See main text for details. In all cases, the first number preceding the name of the subsystem corresponds to its position in the ranking generated by the CV values, which are shown in round brackets here. Context names are displayed in the color bar legend.

Mentions: Alternatively, to further evaluate the functional validity of the RegrEx extracted models, we used the previously calculated MFC to investigate the importance that a given subsystem had in each context. Furthermore, we ranked the subsystems according to the CV of the MFC value distribution of each subsystem across contexts. This implies that subsystems with a low CV are evenly represented among the different contexts, while with increasing CV, these subsystems tend to be more specific for certain contexts. For instance, all subsystems belonging to Central metabolism occupy top positions in the ranking, which is expected due to the fundamental role that these subsystems play in all cell types. The citric acid cycle is the first subsystem in the ranking with a CV value of 0.078. On the contrary, the pentose phosphate pathway is the subsystem in Central metabolism with a highest CV value of 0.21. However it can be considered low in the context of the entire ranking, and may be explained by the fact that, unlike the rest of subsystems in Central metabolism, the totality of its reactions in the core are non-robust, as mentioned before, see Fig 3 and S4 Table. In addition to these subsystems, we also find in top positions others equally fundamental pathways, including: NAD, folate and vitamin A metabolism (all in the category of Cofactor and Vitamin metabolism), extracellular and mitochondrial transport or nucleotides metabolism. Interestingly, the last three subsystems are also the ones containing the greatest number of the previously defined robust reactions (S4 Table).


Context-Specific Metabolic Model Extraction Based on Regularized Least Squares Optimization.

Robaina Estévez S, Nikoloski Z - PLoS ONE (2015)

Illustration of selected Recon 1 subsystems displayed in a Pie chart form, depicting the distribution of mean flux capacity (MFC) values across contexts.Panels A-B correspond to the two extreme metabolic subsystems, in terms of CV, in Central metabolism. The citric acid cycle (A) shows the lowest CV value (both in Central metabolism and within the entirety of Recon 1 subsystems). The pentose phosphate pathway (B) shows the greatest CV value in Central metabolism. (C-E) the distribution of MFC values is shown for fatty acid metabolism (C) which is predominantly represented in liver, adipose tissue and skeletal muscle, fatty acid oxidation (D) and fatty acid activation (E) both subsystems predominant in adipose tissue and skeletal muscle. (F) The MFC distribution across contexts is depicted for gluthatione metabolism. Kidney is the context where this subsystem gets a highest MFC value, constituting a 23% of the total MFC value across contexts. See main text for details. In all cases, the first number preceding the name of the subsystem corresponds to its position in the ranking generated by the CV values, which are shown in round brackets here. Context names are displayed in the color bar legend.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4497637&req=5

pone.0131875.g003: Illustration of selected Recon 1 subsystems displayed in a Pie chart form, depicting the distribution of mean flux capacity (MFC) values across contexts.Panels A-B correspond to the two extreme metabolic subsystems, in terms of CV, in Central metabolism. The citric acid cycle (A) shows the lowest CV value (both in Central metabolism and within the entirety of Recon 1 subsystems). The pentose phosphate pathway (B) shows the greatest CV value in Central metabolism. (C-E) the distribution of MFC values is shown for fatty acid metabolism (C) which is predominantly represented in liver, adipose tissue and skeletal muscle, fatty acid oxidation (D) and fatty acid activation (E) both subsystems predominant in adipose tissue and skeletal muscle. (F) The MFC distribution across contexts is depicted for gluthatione metabolism. Kidney is the context where this subsystem gets a highest MFC value, constituting a 23% of the total MFC value across contexts. See main text for details. In all cases, the first number preceding the name of the subsystem corresponds to its position in the ranking generated by the CV values, which are shown in round brackets here. Context names are displayed in the color bar legend.
Mentions: Alternatively, to further evaluate the functional validity of the RegrEx extracted models, we used the previously calculated MFC to investigate the importance that a given subsystem had in each context. Furthermore, we ranked the subsystems according to the CV of the MFC value distribution of each subsystem across contexts. This implies that subsystems with a low CV are evenly represented among the different contexts, while with increasing CV, these subsystems tend to be more specific for certain contexts. For instance, all subsystems belonging to Central metabolism occupy top positions in the ranking, which is expected due to the fundamental role that these subsystems play in all cell types. The citric acid cycle is the first subsystem in the ranking with a CV value of 0.078. On the contrary, the pentose phosphate pathway is the subsystem in Central metabolism with a highest CV value of 0.21. However it can be considered low in the context of the entire ranking, and may be explained by the fact that, unlike the rest of subsystems in Central metabolism, the totality of its reactions in the core are non-robust, as mentioned before, see Fig 3 and S4 Table. In addition to these subsystems, we also find in top positions others equally fundamental pathways, including: NAD, folate and vitamin A metabolism (all in the category of Cofactor and Vitamin metabolism), extracellular and mitochondrial transport or nucleotides metabolism. Interestingly, the last three subsystems are also the ones containing the greatest number of the previously defined robust reactions (S4 Table).

Bottom Line: Here we present RegrEx, a fully automated approach that relies solely on context-specific data and ℓ1-norm regularization to extract a context-specific model and to provide a flux distribution that maximizes its correlation to data.Moreover, the publically available implementation of RegrEx was used to extract 11 context-specific human models using publicly available RNAseq expression profiles, Recon1 and also Recon2, the most recent human metabolic model.Therefore, our study sets the ground for applications of other regularization techniques in large-scale metabolic modeling.

View Article: PubMed Central - PubMed

Affiliation: Systems Biology and Mathematical Modeling Group, Max Planck Institute of Molecular Plant Physiology (MPIMP), Potsdam-Golm, Germany.

ABSTRACT
Genome-scale metabolic models have proven highly valuable in investigating cell physiology. Recent advances include the development of methods to extract context-specific models capable of describing metabolism under more specific scenarios (e.g., cell types). Yet, none of the existing computational approaches allows for a fully automated model extraction and determination of a flux distribution independent of user-defined parameters. Here we present RegrEx, a fully automated approach that relies solely on context-specific data and ℓ1-norm regularization to extract a context-specific model and to provide a flux distribution that maximizes its correlation to data. Moreover, the publically available implementation of RegrEx was used to extract 11 context-specific human models using publicly available RNAseq expression profiles, Recon1 and also Recon2, the most recent human metabolic model. The comparison of the performance of RegrEx and its contending alternatives demonstrates that the proposed method extracts models for which both the structure, i.e., reactions included, and the flux distributions are in concordance with the employed data. These findings are supported by validation and comparison of method performance on additional data not used in context-specific model extraction. Therefore, our study sets the ground for applications of other regularization techniques in large-scale metabolic modeling.

No MeSH data available.