KITP Qbio recap and thoughts
This post will be a place for me to dump my thoughts about things I did/learned at the KITP 2017 Qbio course, especially as it relates to my research.
Ecological data analysis
One specific theme that kept arising in the talks, especially those about real microbial systems was the difficulty in analyzing ecological data. It seems to be hard to infer things from these metagenomic datasets.
There were a couple of talks which touched on this; Otto Cordero’s work1 seems to use a lot of analyses of entropy. Joshua Weitz had some stuff on trying to explicitly model time series data with short sampling2; however, I think this is similar to stuff already done on the microbiome which usually ends up uncompelling. Aleksandra Walczak has some new stuff3 on analyzing things for similar habitat dependence which I haven’t read yet but looks interesting.
Overall, I’m still not convinced that we can tease apart interaction in complex ecosystems where a bunch of guys are going to matter. We may be able to do so when a lot of types have similar interactions (eg when we can tease apart coarse metabolic pathways). I think getting some kind of sense of similarities between different guys (like in the Tikhonov sub-OTU paper) could be a way to proceed; maybe doing some sort of clustering/PCA type thing on the time series? Another thought would be to analyze time series like people analyze natural language data - give two types a similarity score based on whether or not they are associated with the same sets of other types. Still a lot to think about here.
I also befriended a marine virologist, Gur Hevronig, who is a grad student at the Technion. He’s working on a project to understand the dynamics of bacteria and phages in the ocean. He has a really cool dataset with bacterial and phage abundances for 2 hour intervals in the ocean at different times of year. I may try to do a group meeting presenting his data to try to brainstorm different approaches to understanding that type of data.
Ecology theory
There were a couple of talks on theoretical ecology, both of them focusing on resource-consumer models. Pankaj Mehta talked a bit about some cavity method calculations in MacArthur resource models4, is working on a model with cross-feeding. Ned Wingreen talked about a model with tradeoffs which lets an arbitrary number of guys survive on a fixed number of resources5. However this model requires some strong linearity assumptions, which in general should not be true. Might be interesting to think about perturbations of that model which have long decay timescales. There was also some talk that ecological systems could self organize to that sort of state but I don’t really understand those things to be honest.
I think the main takeaway I came with from talks on the actual biology with regards to directions for theory was the importance of metabolism. There were quite a few talks about metabolism, and many people working in the field seem to think that this will be the only reasonable way to understand the complexity in nature. Alfred Spormann gave a couple of lectures on this in simple systems, and Otto Cordero espoused this view in both of his talks.
It seems that for the systems presented (eg Rachel Dutton’s cheese model community), the first steps to understanding will be based on the large abundance taxa which are doing things like adjusting pH, breaking down complex carbohydrates, and producing antibiotics. The “statistical physicist’s view” of analyzing a large number of types may not be a useful idea here, where the particulars of the environment tell us a lot about what may or may not happen. This does need to be taken with a bit of a grain of salt though; right now people are trying to work on communities chosen for their apparent simplicity. Not sure if one needs to think about a bunch of different simple communities or one complex one. The truth is both approaches are going to be useful for different things.
It might be useful to think about metabolism and tradeoffs, and see if there are some ways of getting the flavors of these processes into models. Not sure how to do this - maybe with some resource model that has a cascade of energy scales, and seeing what can evolve on top of that? Another thought would be to try to see what metabolic relationships/tradeoffs would say about the statistics of interactions.
The question of diversity may almost be something done on a “background” of metabolism/trophic levels/etc. There is certainly some structure coming from these different metabolic pathways, and the presence/absence of different species set the availability (and sometimes the existence) of different resources. However the diversity seen in nature seems to be beyond this, so while understanding the metabolic flows might be enough for a coarse understanding the finer scale details may be independent (or at least less determined by) the broader metabolic flows. An interesting question would be how this fine scale diversity feeds back into the coarser structure as it’s shaped by evolutionary dynamics.
Bacteria-phage dynamics
There were a few talks on bacteria-phage dynamics, a model of which I’ve been thinking about recently. Bruce Levin talked about some of the specifics of viral biology, going into the differences between lytic phage (those that just reproduce and burst the bacteria right away) and lysogenic phage (which incorporate themselves into DNA, and bide their time before inducing replication). He also talked about some experiments in chemostats where bacteria evolve phage resistance pretty quickly. Paul Turner talked about the use of phage in therapeutics, where they might be easier to “design” against bacteria (ie by isolation and experimental evolution), and can be used to both directly reduce bacterial number and more importantly induce tradeoffs that make them less harmful/easier to hit with antibiotics.
Joshua Weitz talked a bit about analyzing data from ocean phage6 and modeling bacterial-phage coevolution. Some papers of his are in the reading list below. There is some claim that stability in these models is helped by having some sort of block structure to the interactions; I need to go through and look at the modeling in more detail to figure out what is going on there.
Overall the basic facts about bacteria-phage systems, things like \(P<B<P+R\) (where \(P\) is phage, \(B\) is bacteria, and \(R\) is resources), the fact that phage generally cause oscillations/etc, seem to be well known; however, considering what happens when you have lots of phage and bacteria has not really been done, nor have questions about co-evolution been satisfactorily been answered.
While it isn’t about bacteria-phage coevolution, Aleksandra Walczak has some work on immune system coevolution7. Might be worth looking at for inspiration.
Random tidbits
There were a lot of other random talks not fitting into the above themes that I thought were interesting. A non-comprehensive list:
- Phillip Messer gave a great talk on trying to understand the evolutionary dynamics of CRISPR based gene drive technologies. A cool look into “applied” pop gen.
- Terry Hwa talked about some work to understand motility in E. Coli, and had some cool data/modeling showing some traveling waves of bacteria spreading on plates through a combination of growth and gradient generated motility. He also had a great lunchtime talk about his stuff on proteome allocation, which had some good philosophical thoughts about research in quantitative biology.
- Paul Rainey had kind of a cool talk on REPINs, these really small genetic elements which may be selfish elements which nonetheless may act as a sort of primitive CRISPR system.
- There were a couple of talks on ideas of group selection and the evolution of altruism, though I didn’t find any particular one compelling in terms of the modeling. Still was good to be exposed to thoughts in that area, and it might be a fun thing to study sometime.
- Aleksandra Walczak talked about some immunology stuff, and I think she gave a second talk the week after the course. Haven’t seen the second one but the first one was quite nice.
Genomics - data analysis and genome structure
I’ll conclude with some thoughts on technical things I learned during the course. My main goal was to get more familiar with analyzing genomics data, which I did get to do to some degree.
Got some exposure to doing bioinformatics type things. Did simple things like mapping reads to contigs, making pileups, etc. Gives me more confidence in doing some of the basics of analyzing sequencing data.
I also got some more exposure to various properties of genome structure. Things like plasmids, transposable elements, codon bias, distributions of genetic elements. May be worth sitting down and learning some genomics properly sometime, though not sure if it would be better to read some sort of papers/reviews, or read through a textbook.
I think I still don’t have enough of a sense of the properties of genomic data analysis/genomes themselves. For many of these things I don’t have a good quantitative sense about when certain things matter, or what their significance is.
Things I should try to learn at some point:
Details about sequencing. I learned a bit about the format of files coming out of sequencing, and some basics about tools used to map/annotate. I still don’t have a good sense of some of the numbers involved, and what the properties of the different technologies are.
Facts about genome structure. Even things like GC bias were new to me; I think there are quite a few things I should know about analyzing sequencing data. Things like frequency of different types of errors, probabilities of spurious matches given fragment lengths, etc. Should maybe talk to Mike or Lily about this sort of thing at some point. Specific questions:
- What sort of codon bias/GC content bias is observed across different types? When is a difference in GC content actually significant of something?
- What kinds of error rates are typical in various sequencing technologies? What kind of forms do those errors take?
- How easy/hard is it to get DNA from one organism to map onto that of another? How closely related do they have to be? Is there some way to get a sense of false positives?
- What does coverage bias across a genome look like? What sorts of regions are conserved enough to give lots of spurious hits? How can those regions be identified/filtered out?
- How much can contigs from assemblies be trusted? How different is this for metagenomic samples versus ones from clones?
- How good are abundance estimations from metagenomic samples? How does this depend on what sorts of databases get mapped to? How does it depend on which genes are actually being picked up?
Knowledge of bioinformatics tools. Ended up getting some exposure to things like assembly (spades), mapping (bwa and fr-hit), annotation (mg-rast,diamond), and data analysis/visualization (MEGAN). Might want to get a sense of what other major players there are out there, so I don’t constantly reinvent the wheel.
Better understanding of bioinformatics algorithms. I learned a bit about the various algorithms, such as the k-mer search approach that seems to be popular. Even then, it seems that things like assembly involve some fiddling around with different parameters to get things to work out. Some thinking about these issues would be good, as would getting a handle on some of how different properties of analyses scale with parameters/change with algorithms.
Reading list
Bacteria-phage
Multiple infections and spatial structure
Phage-bacteria interaction networks
Inferring interactions from time series
Nested infection network model
Old E. Coli phage modeling paper
Genome structure
Model of metabolic network evolution
Early paper on clustering (not sure if good)
Microbial ecology
Review of community interactions
Species variance driven by habitat in microbiome
Ecology modeling
Cavity method in resource models
Renyi entropy as diversity measure
Immunology
Well adapted immune organization
Evolutionary dynamics of immune system
Cell biophysics
Terry Hwa proteome allocation work:
Fun
REPINs (small transposable elements)
Minimum entropy production principle