Recent Changes

Re: False discovery rate — by Niko Balkenhol — last modified 2009-03-16 19:47

Sorry - looks like the links to the two FDR papers aren't working properly :O(

Re: False discovery rate — by Niko Balkenhol — last modified 2009-03-16 19:46

On second thought: If you include FDR, you should perhaps also cover (sequential) Bonferroni correction. I’m a little confused about the separation of AIC vs. Information-theoretic approaches.  After all, AIC is one (of many) measures used in information theory.  I’m also wondering whether you could discuss why using AIC and related indices might not work with our data.  (I.e., why EXACTLY is pair-wise data problematic for this approach? – I took a 3-day workshop on information-theoretic approaches, but even the experts couldn’t really tell me what conceptual/ theoretic arguments go against it for our typical data in landscape genetics…curious to see what you have to say about it…)   This would be a good lecture to point out the problems associated with partial Mantel tests.   And, just a very minor thing: On slides 55ff, you talk about the “correct” model.  All models are wrong, but some are useful.  Maybe change it to “model most supported by the data”?  

pop size & genetic differentiation — by Niko Balkenhol — last modified 2009-03-16 19:44

Hi Rolf,   You might already point it out, but effective population size can also influence neutral genetic differentiation, not just neutral genetic diversity.  (In fact: Are you explaining the difference between diversity and structure, or is someone else?)  I feel that’s often overlooked in current studies…we mostly focus on the landscape in-between samples, but location-specific factors can also influence gene flow (e.g., density-dependent dispersal, reproduction and survival).  Thus, if we have populations of varying sizes/densities, we might expect certain genetic structures, even if gene flow is not influence by space or the matrix.  (And this might actually tie in niceley with the net-work approaches discussed by Melanie in lecture 9.)   Looks great otherwise!

Re: Link to lecture 9 — by Niko Balkenhol — last modified 2009-03-16 19:41

Hi Helene,   This lecture looks awesome – I especially like the mix of theoretic practical considerations (software etc.).  When you mention graph-theory, you might want to cite Conefor Sensinode or related papers for completeness: http://www.conefor.udl.es/

additional citations — by Niko Balkenhol — last modified 2009-03-16 19:40

Hi Stephanie,   Your lecture looks great!  I only have a couple of minor suggestions:   You might want to point out some more that edge detection and clustering methods have similar goals in genetics, even though they are analytically different.  Perhaps we should think about combining these complementary analyses more often.  (Maybe cite Jacquez, G.M., Kaufmann, A., Goovaerts, P (2008) Boundaries, links and clusters: a new paradigm in spatial analysis? Environmental and Ecological Statistics 15(4): 403-419.) I think it’s important that students in the course realize that they should always use more than one program, for example one spatial and one non-spatial clustering method.  It’s also crucial that we start to report exactly how we interpreted outputs of these programs and what decisions we made to derive final conclusions.   For the sampling effects on Structure results, you might also want to cite: Schwartz MK, McKelvey KS (2009) Why sampling scheme matters: the effect of sampling scheme on landscape gnetic results. Conserv Genet. doi:10.1007/s10592-008-9622-1   Finally, on slide 47, where you suggest that assignment probability can be linked to the landscape, you might wan to cite: Murphy, M.A., Evans, J.S., Cushman, S.A., Storfer, A. (2009) Representing genetic variation as continuous surfaces: an approach for identifying spatial dependency in landscape genetic studies.  Ecography 31: 685-697.

Re: False discovery rate — by Niko Balkenhol — last modified 2009-03-16 17:23

I'm not sure which FDR papers Stephanie has in mind, but here are 2 of the main FDR references I'm aware of: Benjamini Y, Hochberg Y (1995) "Controlling the false discovery rate: a practical and powerful approach to multiple testing." Journal of the Royal Statistical Society B 8 57, 289-300.   Benjamini Y ,Yekutieli D (2001) "T he control of the false discovery rate in multiple testing under dependency ". Annals of Statistics 29 (4): 1165–1188.

Re: random vs. IBD — by Niko Balkenhol — last modified 2009-03-16 17:09

I'm think I agree with both of you.  IBD-like patterns can arise from purely spatial influences on gene flow, but also from environmental heterogeneity.  Thinking about the underlying processes, panmixia should lead to spatial-genetic randomness, because all possibe pairs of mating partners in the study area have an equally likely chance of actually mating (i.e., successful mating between individuals is not dependent on their relative locations).  Thus, panmixia is a null, and one might consider IBD a deviation from this null, because it arises in situations where successful mating is more likely among individuals that are close in (environmental) space.  On the other hand, panmixia is not a realistic null-model in an ecological sense, because it is highly unlikely that mating in natural populations really is spatially random.  (Aside fom the fact that space use of related individuals can really mess up our statistical analysis).  Thus, we will very often conclude that our data does not conform to the "expected null", simply because the null is only valid in theory. So, maybe we need to distinguish between analytical and ecological null models?

Re: Link to lecture 6 — by Niko Balkenhol — last modified 2009-03-16 16:40

Hi Melanie, Yes, we'll actually use the "panmictic" scenario as a null in our lab, but that's probably not clear yet.  Thanks for pointing it out...

Re: Simulations for theoretical expectations — by Niko Balkenhol — last modified 2009-03-16 16:38

Excellent point - the Epperson stuff covers this a little bit, but the OIKOS study is even better...I'll think about including it!

Re: Overlap between lectures 8 and 11. — by Helene Wagner — last modified 2009-03-15 18:59

Yes, we'll figure out which of the slides go in which lecture once the final order of lectures has been decided. Yes, the paper should be assigned once (early). Helene

Re: move up? — by Melanie Murphy — last modified 2009-03-15 17:23

I agree - I think it would fit well after lecture 3 or after delineation of populations/groups.  It seems to fit well in the genetic tools section.

Re: Other sequential organization for lectures? — by Melanie Murphy — last modified 2009-03-15 15:26

Hi All - I like Lisette's idea - I think it is a good idea to introduce the landscape concepts early.

Re: minor english errors — by Stephanie Manel — last modified 2009-03-15 12:28

Dear Lisette, I send by e-mail my resentation. It was big. If here is problem I will try again tomorrow morning. About assumptions, not sure to understand what do you mean. Stéphanie

Re: Assumptions — by Stephanie Manel — last modified 2009-03-15 12:25

Dear Melanie ABout slide 45, in fact at point 0: there is no booudary and succes is measrured as no boundary.  AT generation 100 and after, pourcentage of success is detection of boundary. The value of 0 means that no boundary have been detected. I hope that it is clear. I probably need to add the explanation on the slide. Concerning the spatial question: we need to discuss more about that together. See you soon Stéphanie

move up? — by Lisette Waits — last modified 2009-03-15 11:02

rodney, I think this will go better after lecture 3..... Do you agree? This covers the important plant specific methods...I think I should add a more "animal" centric section at the end that discusses how trapping/non-invasive genetic sampling can be used to pick up movement of animals and that assignment test approaches can be used to detect current migrants and offspring of migrants (since pedrigee methods can rarely be used).  Lisette

Re: Debate about partial mantel tests - covering and appropriate lecture — by Lisette Waits — last modified 2009-03-15 10:56

I agree...I think we should address the debate the first time that the method is introduced...so the best location will depend on the final order of presentations. Lisette

Re: Overlapp — by Lisette Waits — last modified 2009-03-15 10:55

I think it is good and the students need to be reminded. Lisette

minor english errors — by Lisette Waits — last modified 2009-03-15 10:53

Stephanie, the content is great...there are  a few minor english errors scattered throughout the slides...if you want to send me the powerpoint file, then i could quickly correct them and send it back to you. i agree with Melanie that it is important to add information about the assumptions of the methods. Lisette

Re: Sampling design — by Lisette Waits — last modified 2009-03-15 10:51

I agree with Melanie that a general sampling design recommendations are not possible because it depends on the question.  But it is true that I didn't introduce the topic at all and it is important and needs attention.  I'll think about how to introduce it and then others can continue the theme. Lisette

Re: Spatial autocorrelation — by Lisette Waits — last modified 2009-03-15 10:48

It is used in the gene flow discussion paper that I was suggesting we read for this week so I wanted to introduce the concept.  I'm not all that knowlegdable about the spatial autocorrelation stuff or Moran's I so if it fits in a later lecture that would be great...i could drop it from here.  But it will make the discussion paper more difficult to follow or I could try to think of a different one.  What did folks think of this student activity?  With more time I could design a better activity that analyzes a dataset.  I had considered having students calculate genetic distance measures - Fst, Ds, Dsp etc and having them interpret the results.  We also still need a good background reading on measuring gene flow.... Listte

Re: Genetic distance measures — by Lisette Waits — last modified 2009-03-15 10:42

I just finished reading thru all the other slides so yes, I'll add the genetic distance measures and a slide about assumptions (if you already have something made up let me know).  concerning the assignment test stuff.  are you suggesting that it goes in Stephanie's defining populations or Dyer's measure direct gene flow?  I was thinking if we moved Dyer's lecture after my gene flow lecture as you and I are suggesting...he could start with the plant specific stuff and I can then end with assignment tests as the more commonly used "direct gene flow" method with animals...then that would flow right into Stephanie's defining populations.  Since assignment tests were orginally applied to defined populations I thought that simplier concept should be clearly introduced before we get to defining populations using bayesian clustering.  And stephanie has a whole lot to cover in her lecture. What do you think of this proposed change? lisette

Re: Citation of figures, graphics, and pictures — by Lisette Waits — last modified 2009-03-15 10:34

Good point!! Lisette

Re: Missing topic - genetic point data — by Lisette Waits — last modified 2009-03-15 10:33

I briefly touch on # 1 and #2 in the gene flow lecture week 3...I'll think about whether I can emphasize more or make it more clear...perhaps it fits best there?  However, that lecture is really full of different methods and might already be overwhelming....some we never come back to like AMOVA so perhaps could be dropped. Lisette

Re: Overlap with lectures 8, 11, & 12 — by Lisette Waits — last modified 2009-03-15 10:31

I agree with this...i think the overlap can be decreased. Lisette

Re: Missing topic - Spatial regression models — by Lisette Waits — last modified 2009-03-15 10:30

could go in the new and hot part??  Ideas are older but use in landscape genetics is new. Lisette

Re: Extra topic suggestion: Difference among genetic markers — by Lisette Waits — last modified 2009-03-15 10:29

Hi all, I think this is covered a bit in lecture 2 and I suggested a background reading on this Sunnuck 2001 that I uploaded so that there could be an extra discussion group on this topic at each university if needed.  I will take another look at my lecture 2 material and try to emphasize it more because you are right that it is a key concept. Lisette

Re: Other sequential organization for lectures? — by Lisette Waits — last modified 2009-03-15 10:26

Hi All, I was also thinking thru the order and came to a similiar conclusion...I favor melanie's suggestion above but I also wonder if lecture 5 should move to lecture 2....I am imagining that each group will spend day 1 discussing the review papers on What is landscape genetics.  It seems like having a lecture right after on What is a landscape would be good...then What is genetic diversity/gene flow etc...that seems more integrated than our current layout. Lisette

Re: Lecutre 2 comments — by Lisette Waits — last modified 2009-03-15 10:14

thanks!  I'll take care of your suggestions. Lisette

Link to lecture 5 — by Melanie Murphy — last modified 2009-03-12 20:39

Hi Sam - There is a nice opportunity in this lecture to make a link to lecture 5 (components of scale, for example). There is a scaling component in the network lecture, so maybe this lecture should come first?  If this lecture is moved earlier in the series, I can also link to it in my network lecture (scale of variable and bandwidth as two additional approaches to scaling).

Overlap with lectures 8, 11, & 12 — by Melanie Murphy — last modified 2009-03-12 20:21

Lecture 8, 11 and 12 all cover vital concepts in landscape genetics.  However,there seems to be a lot of overlap among lectures 8, 11, and 12.  What concepts should be covered in each of these lectures?  Are they easily seperable or should they be condenced into two lectures?  Are there additional examples that could be added to reduce the redundancy?

Missing topic - Spatial regression models — by Melanie Murphy — last modified 2009-03-12 20:17

Should we include anything on spatial autoregressive models (SAR, CAR, STARMA, STAR)? 

Missing topic - genetic point data — by Melanie Murphy — last modified 2009-03-12 20:15

I am not sure where this would fit, but seems like we should mention use of genetic data as point data (and also introduction of point pattern statistics).  For example: 1) distribution of mtDNA haplotypes in relation to landscape features/process 2) spatial distribution of allele(s) 3) identification of cryptic species and their spatial distribution (or non-cryptic species using non-invasive methods) 4) Genetic markers used  to identify disease These methods are not used frequently (yet) and with everything to cover may not be worth > time.  However, I think it would be beneficial to at least have a slide mentioning differnent ways to use genetic data.

Re: Other sequential organization for lectures? — by Melanie Murphy — last modified 2009-03-12 20:07

I think this is a really good suggestion.  I would add that the bear example in 8 (and other least-cost path papers published in landscape genetics) use model selection.  I think lecture 8 could also benefit from being after lecture 11 as well.  It is important to introduce the landscape components (which are half of equation) early in the seminar so participants are thinking about landscape ecology as we go through the semester.  Finally, the scale decisions presented in lecture 12 are applied in lecture 8 (if memory serves).  So maybe: 2 3 4 13 5 6 7 11 12 8 9 14 10 (?)

Overlap between lectures 8 and 11. — by Melanie Murphy — last modified 2009-03-12 19:52

The N Idaho bear dataset is great and should definitely be a central case study in one of the lectures.  In addition, there are some really fantastic slides associated with this dataset. There is a lot of overlap between exercises 8 and 11 realated to this dataset (slides ~13-30, 44-50 in lecture 11 are also in lecture 8).  Also the final results are presented in both lectures. Couple of ideas that might help reduce redundancy: 1) Cover causal modeling more generally, or with examples from ecology in Lecture 11. 2) De-emphasize the Cushman et al paper in lecture 8 and bring out some of the other least-cost path/distance approach papers in landscape genetics as examples. 3) It would likely be way too much information, but collapse 8 and 11 into a single lecture opening up the slot for other topics. It also seems that the Cushman et al paper should be read as the reading assignment associated with the first time the dataset is discussed, and not later in the serries.

Mantel Test/Partial Mantel test — by Melanie Murphy — last modified 2009-03-12 14:21

At some point in the seminar, the debate about partial mantel tests and randomization tests should be covered.  Would this be a good place for this conversation?

Deriving costs — by Melanie Murphy — last modified 2009-03-12 14:20

Hi Sam - Are you going to spend any time talking about how to derive costs/resistance values? 

Re: Some comments on distances — by Melanie Murphy — last modified 2009-03-12 14:17

Hi Sam - The distances would be a good opportunity to link to lecture 3 where these are introduced.  Fst is heterozygosity based (and pop) and next go on to allele based, individual measures.   It is a little apples and oranges.  Maybe introduce a population based distance measure that is allele similarity/frequency based (Nei's D, genetic chord, etc) before going on to individual based measures. 

Link to lecture 6 — by Melanie Murphy — last modified 2009-03-12 13:49

Hi Niko - In the coverage of your simulation experiment, I think you could make a nice link to the null models lecture (#6) and the importance of selecting a null.

Re: False discovery rate — by Melanie Murphy — last modified 2009-03-12 13:46

Hi Stephanie - Could you post the reference for this paper for all of us?  I am very interested in this paper, and I am sure others are as well.

Debate about partial mantel tests - covering and appropriate lecture — by Melanie Murphy — last modified 2009-03-12 13:35

The last part of this lecture deals with mantel/partial mantel tests.  As some point in the seminar serries we should cover the debate over the partial mantel test, appropriate null distributions, and how this relates to multiple matrix regression.  I am not sure that this lecture is the appropriate place, but seems like it might be useful to cover it before the  partial mantel/mantel are introduced.

Years vs. Generations — by Melanie Murphy — last modified 2009-03-12 13:33

Hi Erin - Nice presentation of your simulation approach. Just a small comment, but on slide #37 you list one of the simulation conditions as "years or generations".  This is only years if 1 year = 1 generation, right?  So maybe less confusing to just say generations.

Simulations for theoretical expectations — by Melanie Murphy — last modified 2009-03-12 13:31

Dear Niko, Erin and Sam - Very nice lecture with good examples on how simulations can be used in landscape genetics.  Another use (that you allude to) is using simulation to get a handle on theoretical expectations (from a pop gen standpoint) with landscape effects.  One slightly older reference that focuses on the population genetic effects of landscape processes is: Ezard, T. H. G., and J. M. J. Travis. 2006. The impact of habitat loss and fragmentation on genetic drift and fixation time. Oikos 114:367-375. It might be worth a passing mention, even if not focused on in the lecture, just to make participants aware. 

Re: random vs. IBD — by Melanie Murphy — last modified 2009-03-12 12:55

Dear Helene and Rolf - Nice set of slides with good explanation of underlying assumptions in population genetics and how to think about in them in a landscape context.  I think we should be thinking of randomness as the true null.  It is true that spatial structure has to be present in order to model landscape process (if there is not structure, then this is not possible).  So an underlying assumption with a landscape genetics study is that there is some spatial structure in the data.  However, distance (as a process) is not required in order to get spatial structure.  IBD implys that the resulting pattern (or some portion of it) is actually the result of some distance die-off. Especially at fine scales, there may be spatial structure that is not the result of distance limitations.  What I mean is that without some landscape feature/process, the study area would be random (panmixia).  If we were to test IBD, we would get a significant result (assuming that the landscape feature(s) are autocorrelated).  But this result is misleading ("false autocorrlation", Legendre et al 2002) in that distance among individuals is not actually the process generating the observed pattern.  We demonstrated this in our Ecography paper (Murphy et al 2008) and I believe Niko saw the same result in his Ecography in press paper. If we see spatial structure, it could be the result of one of three general hypotheses: 1) distance (distance die-off in probability of migration), 2) landscape feature(s)/process or 3) distance and landscape.  Although it is very likely and logical that distance is a factor, assuming that is must be present (as the actual generating process) in order to have landscape genetic structure is potentially misleading.

Link to lecture 9 — by Melanie Murphy — last modified 2009-03-11 21:14

I will build a link to lecture nine, reinforcing the graph theory connection. 

Assumptions — by Melanie Murphy — last modified 2009-03-11 21:00

Hi Stephanie - Nice coverage of different methods that are available laced with some excellent examples.  It looks like all of the methods fail at 100 generations (slide 45).  Is this a lack of power issue? I am sure you will talk about this in the presentation of the material, but very interesting. I am concerned about the idea of using spatial assignment tests for further spatial analysis.  In several of the spatial approaches, they could impose spatial structure on the data.  If those data are then analyzed in a spatial context, it could lead to false inference. 

Genetic distance measures — by Melanie Murphy — last modified 2009-03-11 20:38

Hi Lisette - I mentioned this in one of the other comments, but it would be extremely helpful to have a slide/table that talks about the assumptions of genetic distance measures (HW, mutational model, etc).  Also, it looks like both Sam and I will be talking about proportion of shared alleles.  It can be calculated at both pop level and individual level.  Would it be too much trouble to add it to the lecture?  Dps does not make any assumptions of HW or linkage eq, so is useful in cases of recent population change.  I can explain it in my lecture if needed, but seems appropriate to mention it here. For alleles in space, is it worth-while to mention how spatial locations area assigned to the midpoint of the vector connecting sites? I think assignment tests fall nicely into the next lecture.  Maybe give a brief introduction alluding to the next lecture?

Re: Spatial autocorrelation — by Melanie Murphy — last modified 2009-03-11 20:02

Maybe in this lecture (or as part of another lecture) it would be appropriate to discuss a set of methods that are spatial, but do not include processes other than distance (Moran's I, Correlograms, NN, etc).  Was the intention of "distance based measures" to be genetic distance (and include landscape variables in modes) or more purely spatial measures ("spatial genetics" or "geographical genetics")? 

Re: Sampling design — by Melanie Murphy — last modified 2009-03-11 19:42

Maybe we can introduce the general idea here and then come back to it again in later lectures.  The sample design will vary depending on the question and analytical methods.  In addition, unsampled populations have an effect on the interpretation of the genetic data.  However, different spatial methods are more/less sensitive to missing sites (independent of use of genetic data or some other sort of data - networks being particularly sensitive to missing sites).  This would be a great area to point out the need for further research as the interaction between genetic sampling needs and spatial sampling needs is poorly understood. 

Lecutre 2 comments — by Melanie Murphy — last modified 2009-03-11 19:36

Nice introductory lecture to molecular markers.  I really like coverage of HW equalibrium assumptions.  Maybe other lectures can link back to this point (how that effects selection of assignment test/distance measure) and in selected readings throughout the seminar. Couple of graphs could use a citation (I believe slides 25 and 39).  Nicely done discussion questions and would provide nice continuity among sites if questions like these are included with each lecture.  I will update my information to include similar style questions with my selected reading. Exercise - nicely done and good introduction. Would be helpful to add information on why students are selecting options in the program menus.  For example: Explain what the exact test is, and why select that option. Explain what the probability test is.

Re: Link to model selection — by Melanie Murphy — last modified 2009-03-11 17:02

I agree.  I also think it would be good to introduce model selection earlier so people are thinking about it.