The Systems Biology of COVID-19 and the SARS-CoV-2 virus. The class will build a foundation that includes the emergence of complexity, simple biological subsystems, their reductionist and equivalent toy and organ-chip models, and the measurements required to specify model architecture and parameters. Applications to biology, physiology, medicine, chemical and biological defense, pharmacology, drug discovery, and toxicology. UGrad: PHYS 240 01 and BME 290B; Grad: PHYS 326 and BME 395C.
Thursday, April 9, 2015
SysBio15 Asgn_27A_Class_27_Article_22_2015_04_14
Effective Models: Read Article 22 Automated adaptive inference of
phenomenological dynamical models, Bryan C. Daniels and Ilya Nemenman.
Look carefully at all the figures, including those in the Supplemental
Material. Be sure to note the section on the models that don't work.
Post a PCRC.
Subscribe to:
Post Comments (Atom)
Kate Jones
ReplyDeleteAssignment 27A
0. Knew: There is a delicate balance between complexity and simplicity of models--you want a model that is complex enough to predict behavior but simple enough that the model is worth it! We had talked about the dangers of overfitting.
1. Learned: "phenomenological model" and adaptive models. the usefulness of sigmoid models, good models require far less data and time but still accurately predict behavior
2. Pressing ?: Why does overfitting lead to erratic behavior (Figure 3)? I got a little lost on the model selection process in Figure 4.
3. Presentation: Figure 2 comparing different types of models and how each changes with the increased amount of information, Figure 3 showing the effect of overfitting and how the sigmoidal model adapts
4. Thoughts: The analogy of the springs helped me understand how a model that does not reveal any mechanism can still be extremely informative. You don't need to include everything in the model--just enough to predict behavior.
Cameron Togrye
ReplyDeleteAssignment 27A
0. Knew: Overfitting lessens the predictive value of a model. As discussed in previous classes, you want to fit to a certain amount of the data and either A) see if it can predict the rest of the data or B) collect more data down the line and see if it fits that.
1. Learned: The model presented scales the number of parameters (or conversely hidden variables) in order to have the "simplest as possible, but not simpler" model to account for experimental data. The number of hidden variables declines as the amount of data increases, and thus we avoid the overfitting other models suffer from.
2. Pressing: I don't fully understand the network maps in figure 4 part B, and would appreciate explanation there. Also, why does the performance versus computational effort reach a asymptotic limit at .6 correlation in part D of this figure?
3. Presentation: Bayesian Hierarchical Model Selection
4. Thoughts: This paper sort of went a little over my head. I understood the main points it was trying to make, but sometimes got lost in the nitty gritty details. What are the drawbacks of the Sigmoidal phenomenological modeling?
Tim Lee
ReplyDeleteAsgn_27A
0. Knew: An ideal model does not overfit the data nor miss defining features of the system. A good model can predict data trends and be consistent with real data.
1. Learned: Bayesian estimation models, sigmoidal models, and phenomenological models. Different models are better suited for different behaviors of data. Difficulty of fitting the right models is making it as simple and optimal to predict data as possible.
2. Pressing: Most of the supplemental material with explaining the ODE's went over my head. I also do not understand the different conditions in Figure 4 and all of part B in particular.
3. Presentation: Overview of dynamical phenomenological modeling
4. Thoughts: Most of the figures were difficult to understand. It was helpful to see the kinds of behavior data can take on and what modeling has to account for in order for it to be ideal.
Juan Gnecco
ReplyDeleteAsgn 27A
0: issues with overfitting and modeling because its becoming to complex to understand and to lengthy to be efficient.
1: learned their "big picture" approach to avoid being boggled by the microscopic details. and polynomial scaling for dynamic variables.
2: pressing: Obviously got lost with the mathematical approach specially when they got so complicated with the biochemical modeling isn't that just M&M? (4)
3:presentation: Same thing as Cam- the bayesian modeling that system biologists are always talking about.
4: thoughts: supp.fig s2 phosphprylation requires the microscoping details and is missing the spatio. Still not conviced its fully possible to model dynamic systems and rather it is very limited with alot of tradeoffs particularly in biology.
Kendra Oliver
ReplyDeleteAssignment 27A
0. Knew: Now dangers of over fitting a model can occur when parameters are missing from system knowledge. This is done in an effort to replicate and predict a systems but causes the model to be two rigid compared to cellular system.
1. Learned: Using the modeling approach presented in this paper, a increased sense of flexibility allowing for error decrease the amount of data required but are still able to accurately predict biological systems.
2. Pressing ?: Unclear as to mathematical differences in modeling? What exactly is different about this approach? Normal modeling approaches?
3. Presentation: Show benefits of this approach to other modeling approaches.
4. Thoughts: Seems to be nicely incorporating biological instinct with modeling approaches.
Shuaipeng "Jimmy" Zhang
ReplyDeleteAssignment 27A
0. Knew: Some of the basics of model fitting, sigmoidal curves, Hooke's law, law of gravity. Not much else though.
1. Learned: Sir Isaac model, multi-site phosphorylation model, yeast glycolysis model. Classes of phenomenological models.
2. Pressing ?: Difference between S-systems class and sigmoidal models. The article's description of the two was very confusing for me.
3. Presentation: The different types of models introduced in the paper, and their advantages/disadvantages.
4. Thoughts: The ODE's and figures used in the paper were difficult to understand.
Nate Braman
ReplyDeleteAsgn_27A
0. Knew: An overfit model may provide a better match for training data, but will be an ineffective predictor of system behavior under new condition.
1. Learned: Various model examples and their application. The advantages of a good model and the importance of adequate data. Methods of evaluating suitability of model fit.
2. Pressing: Had some trouble interpreting figures. What is the best way of assessing/improving membrane robustness.
3. Presentation: Phenomenological modeling, compared against other modeling approaches
4. Thoughts: Very dense, but interesting article. I'm definitely looking forward to the class discussion, and hope to wrap my head around the math a little better afterwards
Chuck Herring
ReplyDeleteAsgn 27
0: Knew: To be worried about over-fitting and that a better method is needed for modeling.
1: Learned: Their coarse grained approach could be useful
2: Pressing: Need to learn more about the details of their approach, some of the stuff needs time to set in.
3: Presentation: New modeling approach
4: Thoughts: How would you statistically test this model?
Selene van der Walt
ReplyDeleteAsgn_27A
0. Knew: about the basics of most of the examples they used for models such as gravity and multi-site phosphorylation. Also knew about error going down as the amount of data increases.
1. Learned: An adaptive model is one that includes more parameters as it is given more data. Overfitting creates erratic data. When selecting models, the model with the largest connectivity diagram is most likely to make correct out of sample predictions. 'Hidden variables' increase nonlinearity.
2. Pressing ?: I think I am still a little shaky in my understanding of overfitting.
3. Presentation: The process of selecting the best model for your data set, advantages and disadvantages of various techniques.
4. Thoughts: Obviously a very long paper but on an interesting topic, I feel like often models are made without thought to what would actually be the best way to represent that data through a model.
James Pino
ReplyDeleteAsgn 27
0:Knew Modeling practices
1:Learned: This is a good method to construct effective models
2:Questions: Can these models be simulated stochastically?
4:Thoughts: Seems like a great method. I want to see how well it works on such a large model.
Mark Vander Roest
ReplyDeleteAssignment 27A
0. Knew: Basics of models, modeling challenges, fitting challenges.
1. Learned: The application of modelling principles to data on gravity is pretty cool. Not a concept I had thought of before, but it simplified the principles in a way we could understand.
2. Pressing ?: What are strategies for incorporating both the mathematics and biological knowledge required for these types of models?
3. Presentation: Modelling approaches and strategies.
4. Thoughts: Cool topic. Very long format.