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.
Monday, September 26, 2016
Subscribe to:
Post Comments (Atom)
0. Knew:
ReplyDeleteI knew about epigenetic landscapes from the previous papers.
1. Learned:
One thing new that I learned from the model presented is that there exist partially reprogrammed cells that are hybrids of naturally occurring cell fates.
This paper essentially fills in the steps to mathematically formulating the epigenetic landscape and shows application of the model through analysis of data from mouse whole genomics microarrays. The data was assumed to be log-normal distributed, each cell fate was given a z-score, and then the data was discretized for positive and negative z-score values.
Two possible “distance” measurements are introduced to describe landscape results; the first is the overlap (basically a dot product) of the expression state and the gene expression, and the second is the projection of an expression state on a cell fate. The second measurement is better used for cell fates that come from similar lineages.
The overall epigenetic landscape can be viewed as the sum of four terms (equation 4) that accounts for attractor states, bias from experiments, the environment from the culturing conditions, and a low energy path between two cell fates.
Cells that can be reprogrammed in this model should be both highly expressed and highly predictive of the desired cell fate. Therefore, a quantity is defined called the predictivity of a certain TF for a given cell fate. From the definitions, it can be shown that that previously defined projection is just the overlap between the desired expression state and the predictivity of that state.
2. Pressing Questions:
Many quantities were defined in this paper (overlap, projection, predictivity, etc.). I understand what these values are mathematically from experience in linear algebra, but I am still having a hard time relating these quantities to the epigenetics. Which of these values are the most important ones to consider, and what are their significances in this model?
3. Presentation Topic:
A presentation that presents just the mathematical model first and then shows how it is used would be useful. The paper seems to jump between these topics when discussing the results and it would be nice to see them separate first.
Also, the Yamanaka protocol is mentioned a lot in this paper. Would a presentation on that be helpful?
4. Thoughts:
This paper was a bit confusing to me because they jump back and forth between the mathematical model, their own data, results of the model, and previously done settings. As mentioned in the presentation section, I could have followed better with more organization. I know that the whole model is presented later on, but it made reading the main part of the paper a bit challenging.
I agree that it is important to understand exactly why the different quantities were introduced, and which ones are most important. This is not immediately clear, and will require some thought on our part. Probably not much...but a little bit at least.
DeleteKelly McGee
ReplyDeleteArticle 12b
Epigenetic Landscapes Explain Partially Reprogrammed
Cells and Identify Key Reprogramming Genes
0: Knew: From last class' papers, I was aware of attractants and their usefulness (in at least a qualitative sense) in modeling gene regulation networks for cellular differentiation and reprogramming.
1: Learned: I learned of a quantitative model for genetic differentiation regulation networks utilizing attractants. I learned how the authors experimentally tested its effectiveness using "normal" cellular differentiation pathways, and how it explains the results of reprogramming studies.
2: Questions: Have the TF predictions been directly, experimentally confirmed or has proof been made against them since the paper's writing?
Has any more information been added to the algorithms of the model (i.e. genes, more experimental data) to make it more robust since 2014, when the paper was written?
3: Presentation: An example of anything from (2). I would love to see a presentation on anything the model predicts that has since been confirmed or denied.
4: Comments: Interesting paper, and very helpful coming out of last Thursday's more qualitative discussion. No conclusion section?
Class 12. Article 14. A.H. Lang, et al., Epigenetic Landscapes Explain Partially Reprogrammed Cells and Identify Key Reprogramming Genes.
ReplyDelete0. KNEW:
We saw epigenetic landscapes last week in two papers: one which described the idea qualitatively, and one which offered some experimental evidence suggesting it might be a good framework for thinking about cell fate determination. Given that, I understood the basic idea already---but it was great to see a lot more detail on the modeling front on how exactly important transcription factors would be determined, and how different quantities should be computed.
While I am not super familiar with spin-glass physics, I know about spin, and it is not surprising to me that the idea of a two state system is generic enough to be applicable elsewhere.
1. LEARNED:
I learned one way to implement the ideas we discussed last time in a fair amount of detail. The biological input consists of transcription factor expression information obtained from microarray experiments, which is reinterpreted as data some number of standard deviations away (i.e. z-score) from the mean of the presumed distribution. This is done so experiments done in different labs or in different conditions may be compared.
From there, for computational convenience, the continuous z-score data (real numbers between - infinity and infinity) is converted to plus 1s and minus 1s depending on whether the number is above or below zero. Now for each of our 1337 transcription factors, we either have a plus or minus 1. I assume that the z-score=0 case happens infrequently enough that it may arbitrarily be set to either.
The experimentally known cell fates are all characterized in this way. Then some reasonable notion of dynamics is constructed by thinking about how various cell fates overlap; once this has been done, the landscape and dynamics have been completely determined.
From there, you can add extra terms to the 'Hamiltonian' of the model to account for landscape deformations due to a change in external conditions. In other words, things can happen to change the favorability of different cell fates, or of traveling between cell fates, and those can be accounted for in this way.
Random noise is added in, and certain interactions are randomly removed, to account for real trajectories not being deterministic, and transcription factor-transcription factor interactions not being symmetric.
The model naturally accounts for partially reprogrammed states as other, shallower attractors in the landscape (as we discussed before), and gives you a way to figure out which transcription factors might be important for a specific reprogramming to be done. This is done by looking at expression levels, and an introducing quantity called predictivity, and ranking all transcription factors according to them. Both being high suggests the transcription factor should be overexpressed; both being low suggests it should be knocked out.
(cont.)
ReplyDelete2. QUESTIONS:
The whole model seems to be assuming that everything is working under relatively stable biological conditions. That is, cells are free to change, and are not being destroyed or significantly disrupted from changing.
First, then (if I'm right), I wonder if it is problematic to not take this into account. I will say, though, I'm not sure how this would be done. This is analogous to treating a ball rolling down a ramp and trying to account for someone coming in and kicking it off halfway through. Maybe at that point you're dealing with a different problem entirely.
Secondly, and maybe more interestingly, I wonder if we could think of cell death in this way. That is, could we construct a 'cell death landscape' that would be useful in our mechanism of action studies? Different fates (apoptosis, necrosis, etc) would be the natural attractors in this case, and different states could either be thought of as representing certain gene expression levels, or maybe different concentrations of chemicals of interest.
The idea sounds cool to me...but then again it could be completely stupid.
3. PRESENTATION:
It would be nice to see someone go through the different quantities introduced in the model and motivate them, along with discussing their values in some simple boundary cases (i.e. interaction strength when two cell fates are the same or completely different). The qualitative idea has already been presented just fine.
4. THOUGHTS:
Once again I saw math. This time there were things with indices. I'm glad they're there, because it makes me feel like I'm useful...even if they're probably not important enough to dwell on.
Your comment about a 'cell death landscape' sounds very interesting and I would also like to know if that would be a useful way to think about it. It sounds promising to me anyways.
DeleteChinowsky_TheorSysBio_PCRC_100416
ReplyDelete0. Knew
From previous papers, I was familiar with the epigenetic landscape that they speak of early in this paper. I have formerly been exposed to neural networks, but not specifically with their applications in biology.
1. Learned
While this paper is also exploring the concept of an epigenetic landscape, they do so from the combination of neural networks and whole genome expression profiles. The neural networks take a set of vector inputs and construct a global landscape that makes each of the input vectors a global minimum with an attraction basin. It’s surprising that neural networks hadn’t been mentioned previously, as they align so well with the concepts outlined in both of the Huang papers. The “landscape” model that they present does account for a variety of important factors, including research bias towards certain TFs, the importance of the culturing environment, etc.
2. Pressing
The Yamanaka protocol is mentioned in several key places in this paper. What exactly is this protocol and what separates it from other protocols?
3. Presentation
I think it might be beneficial to take at deeper look at the mathematics outlined in this paper, separated from the large body of text.
4. Thoughts
This paper took the concepts that had been built up from the previous papers, and actually presented a quantitative model for an epigenetic landscape based in transcription factors. It was nice to finally “see it all come together” so to speak.
Ben Terrones
ReplyDeleteSysBio16 Asgn_12B_Class_12_Article_14_2016_10_04
0. KNEW:
I knew about the epigenetic landscapes and reprogramming from previous papers.
1. LEARNED:
I learned that a common result of reprogramming is partially reprogrammed cells. These partially reprogrammed cells are hybrids that co-express transcription factors of multiple naturally occurring cell fates. I learned what a content-addressable memory is and how it applies to attractor states in the landscape. The rest of the paper described the mathematics of constructing the landscape, how to measure distances, and how they identified transcription factors for cellular reprogramming. The transcription factor section was particular interesting, specifically looking at predictive ability of certain TFs.
2. PRESSING ?:
What is spin glass physics? This will probably be covered in the class discussion, but the math was confusing at some points and it would be helpful to go over some of the specific equations and ways they came up with them.
3. PRESENTATION:
Spin glass physics and neural networks
4. THOUGHTS:
I liked this paper because after learning about these landscapes, it was interesting to see how they are constructed mathematically. I also thought it was interesting how their model only depends on the experimentally determined gene expression of natural cell fates.
0. KNEW:
ReplyDeleteI was familiar with cell fate, epigenetic landscapes, and attractors from the previous papers. I also knew about the divergence of thought from standard development of cells and the impact that reprogramming cell fate would have. I had been exposed to neural networks, but mostly in the context of machine learning.
1. LEARNED:
I learned that spurious attractors are partially reprogrammed cells which appear stable, but also can express multiple cell fates. I learned about a specific model which tries to recreate the epigenetic landscape, and how this information was tested. I learned about the importance of neural networks in this model as well as the importance of transcription factors.
2. PRESSING QUESTIONS:
My biggest questions come with the math portions of the article, but I tried not too spend too much time dwelling over the details of the paper.
Were there significant differences in the results of the model before the outliers were removed?
3. PRESENTATION TOPIC:
A detailed description of the Takahashi and Yamanaka reprogramming experiments. There are two supplemental articles that were referenced in the paper regarding the findings of the pair.
Also, A closer look at the math behind the simulations.
4. THOUGHTS:
Good change of pace from the last several articles, but was glad it stuck with the common theme.
0. Knew:
ReplyDeleteKnew of the landscapes discussed earlier and how specific cells are "programmed" for one of a handful of end states. Also with the role of TFs in this process.
1. Learned:
Better understanding of partially reprogrammed cells and how epigenetic models can account for them. Also learned some about reprogramming experiments (but would like to know more).
2. Pressing Question:
How exactly were Hopfield Neural Networks used in the derivation of this model?
Also a question about H bias in eqn 4. What biasing by experimentalists is this referring to?
3. Presentation:
Spin-glass physics and/or a the supplementary section on Hopfield neural networks would be beneficial for me
4. Thoughts:
I thought this paper went into a good level of detail and appreciated the number of relevant examples. Including the methods last caused some confusion for me as I couldn't always tell what was provided by the model and what was known prior.
0. KNEW
ReplyDeleteMaterial about epigenetics landscapes covered in other papers.
1. LEARNED
There have already been success in generating neurons, cardiomyocytes, liver cells, NPCs and thyroid cells, which is really impressive. Reprogramming TFs is based on final, not initial, cell fate.
Starting cell fate is forced to express a small number of TF and this leads to a stochastic conversion to the desired cell fate, which only has a weak bias towards the final state. Cells in turn get stuck in a metastable state. During standard development, external signals actively reshape the landscape and has a strong bias to the final cell fate.
2. QUESTION
Are there other ways in which TFs are used to exert an effect on cells and not just to de-differentiate cells?
3. PRESENTATION
A high level overview of the formulas and how they're obtained.
4. THOUGHTS
A detailed and practical paper. I don't have a complete understanding of how the formulas are obtained (it pretty much all goes over my head), but it's nice to see where the logic comes from.
0. KNEW
ReplyDeleteI was familiar with the concept of epigenetics from previous classes. I was also introduced to the concept of an epigenetic landscape from the previous paper. That paper presented the idea of basins of attraction for
1. LEARNED
A striking fact that the paper brings up early on is the capacity of a small set of transcription factors for reprograming cell fate. This happens through the TF’s defining a different trajectory on the epigenetic landscape for cell states to “roll” to another basin of attraction. These partially reprogrammed cells require some external force in order to arrive in such a state which occupies a smaller, spurious basin of attraction. The paper first linked network started in basins of attraction to favorable cell fates.
New information to me also included the fact that cellular identity and differentiation largely rely on epigenetics, especially modification of histones in the genome. Cell fates from similar lineages often express similar GRN states, and so projecting a cell state on a given cell state is often helpful. Also, the paper explains the terms summating to total quasi potential energy, including location on topography, TF bias, culturing conditions, and switch states for driving network state change.
2. PRESSING ?
Unclear to me how overlap and projection differ, for example, in the case of B and T cells which overlap greatly but do not project on one another?
3. PRESENTATION
Yamanaka protocol
4. THOUGHTS
Dense paper…I felt some of the mathematical concepts were not explained fully
Natalie Hawken
ReplyDeleteAsgn 12B, Article 14: Epigenetic Landscapes Explain Partially Reprogrammed Cells and Identify Key Reprogramming Genes PLoS-CB
0. KNEW
From the class's previous papers, I knew about the epigenetic landscape of quasi-potential energy, and I knew how cells fall into basins of attraction in this landscape. I knew that previous experiments have found a set of 4 TFs that lead to reprogrammed cells. I also knew a bit how high-dimensional analysis can be reduced to smaller sets of more encompassing axes.
1. LEARNED
I found it interesting how they distinguished the distance measures for cell types that are very closely related and for types that are very different using different equations. I didn't know that these experiments led to partially reprogrammed cells, and I thought it was helpful that the model could include these cell types. It was interesting to learn why only reprogramming protocols lead to these partially reprogrammed attractor states while standard development only leads to the distinct cell lines. I also learned that the transcription factors needed to produce a certain fate could be applied first to different cell types and still get the same output.
2. MOST PRESSING QUESTIONS
I'm a bit confused as to how the specificity or expression of a TF differs from the predictability of the TF? (pg 8). How is the predictability measure found?
What is a Hopfield neural network?
(Table S1) How many of the top ESC reprogramming candidates have been applied to reprogramming protocol? How well have these TFs worked to reprogram cells?
3. PRESENTATION TOPICS
Turn the equations in table 1 into an explanation of what each variable means and how the equation is applied.
Discuss how the main TFs discussed in the data work at the molecular level to impact gene expression (especially Myc, the less specific TF that still helps efficiency for almost all reprogramming) (pg 8).
4. THOUGHTS
I thought it was really interesting how they used their computational model to plan out future wet-lab experiments. For example, they used expression data for TFs to determine which to overexpress or knockout to reprogram cells.
I wonder if a temporal piece to the TFs is missing. Maybe if the TF's could be knocked out at different time points, the reprogramming process would work better and limit the number of spurious attractor states? Since the cell differentiation process often involved multiple layers of cell types before reaching the final differentiated form.
Sylvia Morrow
ReplyDeleteAsgn12B_A.H. Lang, et al., Epigenetic Landscapes Explain Partially Reprogrammed Cells and Identify Key Reprogramming Genes PLoS-CB
0. KNEW: The basic idea of having an epigenetic landscape and using overexpression to push cells from one type to another.
1. LEARNED: Partially reprogrammed cells are 'hybrid' states thought to correspond to spurious attractors. Primarily this paper was informative in the sense that it gave much more specific details for how the model was constructed and structured than Huang's papers did. Fig. 1 was very informative in how different factors considered affected the landscape. Pg 5 explained noise insertion. They gave clear, experimentally verifiable predictions with some comparison to results. In particular I found this example very helpful in understanding the connection between theory and experiment: "Brachyury (T) [42] is a general marker of mesodermal lineages. Since it is highly expressed in large a number of cell fates, it is not specific to any given cell fate. However, it is predictive because its expression is a strong indicator that a given cell fate is a mesodermal lineage." (pg 8).
2. PRESSING ?:
--I don't understand what Fig. 2E means.
--Would be nice to get a straightforward explanation of what the input biological info was into the model.
--Usually we want to minimize entropy...what does a Maximum Entropy approach imply?
3. PRESENTATION:
--What would be required experimentally to distinguish between their landscape and a weighted interaction network?
4. THOUGHTS: Seemed like a sort of strange paper structure, but overall an informative read.
Stephen Lee
ReplyDeleteClass 12, Assignment 12b
Epigenetic Landscapes Explain Partially Reprogrammed Cells and Identify Key Reprogramming Genes
(0) Knew:
I approached this paper with the knowledge I’ve accumulated from the previous few lectures (epigenetic landscape/attractor states) and Huang’s paper on epigenetics and Post-Darwinian biology. I have background knowledge in determining cell fates from tissue engineering.
(1) Learned:
I learned that cells can co-express genes from multiple cell fates when naturally or artificially (partially) reprogrammed (this was something that had previously been borne out in the data, but not fully explored). I learned more about the mathematics of the landscape model than had been described in the Huang paper (as well as how the biological contexts of this math).
(2) Pressing Questions:
What makes some TFs with high correlations of expression less predictive of cell fate than others? Does it simply have to do with which cells the authors were looking for, or are some simply always co-expressed?
(3) Presentation Topic:
Predictivity vs specificity of TFs (and other genes)
(4) Thoughts:
This was an interesting albeit dense paper. Looking forward to seeing how the discussion proceeds in class.