Combined Week 11 Discussion Questions
Zanin and Boccaletti
Model assumptions and use
“Topological metrics of these networks can be used to easily discriminate ON from control subjects…” Does this type of model only work if you already know which are which? Can this be used to categorize unknown subjects?
“This is equivalent to an eigenvalue problem, where the centralities of the nodes of the network are encoded in the eigenvector associated with the largest eigenvalue of the adjacency matrix. The fit is performed using only the value associated with the most central node, which should be the main responsible of the disease.” Does this assume only one factor responsible for the disease? How would you do this analysis with multiple or combination factors?
Could be the complex network technique useful to the diagnostic of patients that are starting to develop the disease? (Where the levels of metabolites and the gene expression is not so changed)
How can networks influence ON treatment/can it?
The model is based in the relation between metabolites and gene expression. How much do we have to know about the differents regulators of a natural system to build the network?
Clarification
How does the ON complex network differ from an ANN defined network?
What sort of black-box analysis tools are Z&B referring to?
Critiques
10 control and 10 ON does not seem like a very high N to do this sort of analysis. Is it sufficient?
Comparison to other approaches
How does this way of visualizing complex networks differ from other ways of visualizing complex systems? Is it always “added value” because other types of analyses precede the network definition?
Consider neural networks versus genetic algorithms. How might they be used for different purposes? (e.g., data mining, optimization)
Limitations
What are potential pitfalls of using ANN or networks?
Application to Environmental Systems
The authors find that the efficiency of mRNA and metabolic networks can be used to discriminate ON from non-ON patients, with ON patients having more efficient networks. Could network efficiency indicate different states in environment systems? Would degraded systems exhibit different topological structures, with fewer nodes?
Ruddell and Kumar
Model assumptions and use
How does gap filling represent an a more accurate flux estimate? I take it this is effective “smoothing” of the data?
How much of the information that generate the nodes of the network are relevant to develop the process models?
“if the knowledge of a few state variables or coupling relationships is often insufficient to characterize the behavior of the entire system”, How many process do you need to generate a robust model?
How do you deal with the strength and direction of the feedback to develop the network?
R&K mention they can utilize a S/N > 1 to gain significant information– it seems as if aperiodic noise would call for a higher S/N than that, so how do they confirm useful signals?
Clarification
Can you further clarify the types of entropy and the derivative measurements R&K mention?
Data Interpretation
In Figure 6, how can you indentify the direction of information flow?
The authors build on a number of assumptions/hypotheses (which they show are exhibited by their model): coupling strength is strongest when ecosystem actively growing and interactions with climate are strongest; ecosystems of higher order produce the most information; maximum information flow occurs at the developmental peak of the ecosystem; presence or absence of key couplings and feedbacks characterize differences between optimal and sub-optimal ecosystem states. How can we examine/understand these assumptions/hypotheses?
What is the advantage/benefit of distinguishing between different types of couplings?
The ‘‘engine of variability’’ that is necessary for the land surface ecohydrological system to thrive [Kumar, 2007] appears to be broken down during drought because of insufficient information input from the synoptic weather patterns. The moisture fluxes which carry the information may be reduced below a key threshold during drought.” How can we understand variability as an important factor in ecosystem functioning?
The authors note that there is no agreement on how to robustly define drought and that the impact of drought can vary widely. In their model, drought is defined as a system state in a process network. How does this approach affect analyses of extremes and impacts?
Challenges and practical considerations
How do the spatial and temporal resolution of data sets affect how they are characterized using canonical couplings? What re the implications for the exercize as a whole if they are mis-characterized?
How does time scale influence the requirements for this sort of modeling?
Critiques
The example of drought versus normal growing season for the corn-soybean system contrasts feedbacks and coupled processes in the two system states. However, one would not need this statistical method to tell him or her that a drought occurred.
Application to other datasets
“The identified couplings and time scales are evaluated against the prevailing understanding of the system to judge whether the process network thus characterized provides valuable representation.” Can the model then be applied to other (poorly understood) systems without available validation?
How might the type of analysis presented by the authors be applied? (How) do you see it as useful for any of the type of work that you do? What are the drawbacks and limitations?
Can you think of other examples that would be more insightful for ecosystem management purposes? Or, how might the findings about drought feedbacks affect how the system is managed in the future?
Extension to spatially explicit systems
“If feedback coupling breaks down at the regional scale during drought, one should expect increased heterogeneity on the landscape, such that local ecosystems which are fortunate to receive ample rainfall continue to thrive, but are unable to aid surrounding areas via reginal moisture recycling.” How could the analysis presented in this paper fit in with a pattern analysis?
Complex networks analysis of obstructive nephropathy data
The authors find that the efficiency of mRNA and metabolic networks can be used to discriminate ON from non-ON patients, with ON patients having more efficient networks. Could network efficiency indicate different states in environment systems? Would degraded systems exhibit different topological structures, with fewer nodes?
Ecohydrological Process Networks
In Figure 6, how can you indentify the direction of information flow?
How do the spatial and temporal resolution of data sets affect how they are characterized using canonical couplings? What re the implications for the exercize as a whole if they are mis-characterized?
Ruddell and Kumar
“The identified couplings and time scales are evaluated against the prevailing understanding of the system to judge whether the process network thus characterized provides valuable representation.” Can the model then be applied to other (poorly understood) systems without available validation?
The authors build on a number of assumptions/hypotheses (which they show are exhibited by their model): coupling strength is strongest when ecosystem actively growing and interactions with climate are strongest; ecosystems of higher order produce the most information; maximum information flow occurs at the developmental peak of the ecosystem; presence or absence of key couplings and feedbacks characterize differences between optimal and sub-optimal ecosystem states. How can we examine/understand these assumptions/hypotheses?
How might the type of analysis presented by the authors be applied? (How) do you see it as useful for any of the type of work that you do? What are the drawbacks and limitations?
How does time scale influence the requirements for this sort of modeling?
What is the advantage/benefit of distinguishing between different types of couplings?
The ‘‘engine of variability’’ that is necessary for the land surface ecohydrological system to thrive [Kumar, 2007] appears to be broken down during drought because of insufficient information input from the synoptic weather patterns. The moisture fluxes which carry the information may be reduced below a key threshold during drought.” How can we understand variability as an important factor in ecosystem functioning?
The authors note that there is no agreement on how to robustly define drought and that the impact of drought can vary widely. In their model, drought is defined as a system state in a process network. How does this approach affect analyses of extremes and impacts?
“If feedback coupling breaks down at the regional scale during drought, one should expect increased heterogeneity on the landscape, such that local ecosystems which are fortunate to receive ample rainfall continue to thrive, but are unable to aid surrounding areas via reginal moisture recycling.” How could the analysis presented in this paper fit in with a pattern analysis?
Zanin and Boccaletti
10 control and 10 ON does not seem like a very high N to do this sort of analysis. Is it sufficient?
“Topological metrics of these networks can be used to easily discriminate ON from control subjects…” Does this type of model only work if you already know which are which? Can this be used to categorize unknown subjects?
“This is equivalent to an eigenvalue problem, where the centralities of the nodes of the network are encoded in the eigenvector associated with the largest eigenvalue of the adjacency matrix. The fit is performed using only the value associated with the most central node, which should be the main responsible of the disease.” Does this assume only one factor responsible for the disease? How would you do this analysis with multiple or combination factors?
Week 11 discussion questions: Visualizing and resolving process networks (4/2/13)
Ruddell and Kumar, 2009
The example of drought versus normal growing season for the corn-soybean system contrasts feedbacks and coupled processes in the two system states. However, one would not need this statistical method to tell him or her that a drought occurred. Can you think of other examples that would be more insightful for ecosystem management purposes? Or, how might the findings about drought feedbacks affect how the system is managed in the future?
Zanin and Boccaletti, 2011
How does the ON complex network differ from an ANN defined network?
How does this way of visualizing complex networks differ from other ways of visualizing complex systems? Is it always “added value” because other types of analyses precede the network definition?
Consider neural networks versus genetic algorithms. How might they be used for different purposes? (e.g., data mining, optimization)
What are potential pitfalls of using ANN or networks?
M. Zanin and S. Boccaletti
Could be the complex network technique useful to the diagnostic of patients that are starting to develop the disease? (Where the levels of metabolites and the gene expression is not so changed)
The model is based in the relation between metabolites and gene expression. How much do we have to know about the differents regulators of a natural system to build the network?
Ruddell & Kumar
How much of the information that generate the nodes of the network are relevant to develop the process models?
“if the knowledge of a few state variables or coupling relationships is often insufficient to characterize the behavior of the entire system”, How many process do you need to generate a robust model?
How do you deal with the strength and direction of the feedback to develop the network?
R&K mention they can utilize a S/N > 1 to gain significant information– it seems as if aperiodic noise would call for a higher S/N than that, so how do they confirm useful signals?
Can you further clarify the types of entropy and the derivative measurements R&K mention?
How does gap filling represent an a more accurate flux estimate? I take it this is effective “smoothing” of the data?
How can networks influence ON treatment/can it?
What sort of black-box analysis tools are Z&B referring to?
If we were to create a matrix comparing all the modeling methods in this week’s readings, what would be useful/telling to compare? How could we go beyond general pros/cons to capture their applicability, underlying theories and assumptions, classifications, relative uncertainty? Which approaches are deductive vs inductive and what do their respective results reveal about the applicability of deductive vs inductive approaches?
When is it best to employ statistics versus mechanistic process dynamics to model phenomena? How can we compare Ruddel and Kumar’s approach to Larsen and Harvey, both specifically address feedback mechanisms? What about Ruddel and Kumar to Moffat et al, an interesting comparison because they’re both using FLUXNET data?
Ruddel and Kumar’s approach is interesting because it specifically identifies/represents feedbacks (versus forcings or synchronicities), their directionality, and time lag. Clarification: are they measuring the strength/intensity/magnitude of information flow and is that important? How could we apply Ruddel and Kumar’s approach to geomorphic processes or spatially-dependent variables? They provide physical interpretations of findings (on pp 14-15 for instance), but their model isn’t based on physical processes but statistical analysis. Does this make it more/less legitimate?
Do any confront issues of equifinality? They all seem to address/identify hierarchies, but how do they differ and is any one approach more valid/useful than others (i.e. Ruddell and Kumar identify subsystems)? Is shredding an issue in any of these approaches?
In Zanin and Boccaletti, is it appropriate/legitimate to create a network topology based on linear regression analysis of control data (figure 1)? So this method is best for screening, but doesn’t necessarily help identify a cause for the disease? The network topology visualization is telling, could the other ANN/Process Network approaches also create this type of output (as opposed to Ruddell and Kumar’s matrix output…detailed but not graphically compelling)?