[R-sig-eco] Rarefying metagenome data table
Dear List,
I have a large data table (its the TARA Oceans metagenomic data and
can be found here http://ocean-microbiome.embl.de/companion.html).
Essentially the columns of the data table are samples taken in
different parts of the ocean and the rows are different genes found in
those locations. The numbers in the body of the table are the number
of times an insturment detects that gene in a sample. Because the
machine returns more reads (number of observed genes) in some samples
than in others, the data need to be "rarefied", that is subsampled
such that there are the same number of genes assigned to each station.
I am trying to use the rrarefy package in vegan to do this, but I keep
getting an error message
## Get TARA oceans KEGG metagenomes
temp = tempfile()
download.file("http://ocean-microbiome.embl.de/data/TARA243.KO.profile.release.gz;,
temp)
taraKEGG = read.delim(gzfile(temp, "TARA243.KO.profile.release"))
unlink(temp)
## some processing
ids = taraKEGG[,1]
data = taraKEGG[,2:dim(taraKEGG)[2]]
minsamples = min(colSums(data))
mtx= as.matrix(data)
imtx=as.integer(as.matrix(data))
data2 = data
data2 = sapply(round(data, 0), as.integer)
## load library
library(vegan)
## rarify samples
rare = rrarefy(data2, minsamples)
> Error in if (sum(x[i, ]) <= sample[i]) next :missing value where
> TRUE/FALSE needed In
> addition: Warning messages: 1: In rrarefy(data2, minsamples) : Some
> row sums < 'sample' and are not rarefied 2: In sum(x[i, ]) : integer
> overflow - use sum(as.numeric(.))
It looks like the problem is a line in rrarify where it tries to make
a huge matrix that has as many columns as there are genes in one of
the samples
` row <- sample(rep(nm, times = x[i, ]), sample[i])`
Since there are sometimes millions to tens of millions of genes, this
ends up being a really big matrix and the program gives up in order to
save my computer memory.
So, I am trying to figure out how to proceed from here. Perhaps I am
using this function incorrectly and there is a different way to use it
to not have this problem. Alternatively, perhaps I should be using a
different tool. I have heard of rarefication programs in python, or,
for that matter the bioinfomatics universe of Qime, but I'd rather
stay within R if at all possible. Does anyone have any suggestions?
Thank you for your time.
Sincerely,
Jacob Cram
___
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[R-sig-eco] El Nino and GAMM Model of a Seasonal and Long Term Time Series
Hello Gavin and R-sig-ecology list, Based on Gavin Simpson's very helpful advice on this forum, I have been modelling biotic and environmental parameters using a generalized additive mixed model of the form mod - gamm(logit.Area ~ s(DoY, bs = cc) + s(elapsed.days), data = lldata, correlation = corCAR1(form = ~elapsed.days), knots = list(DoY = c(0, 366))) where lldata is provided below and is data for one taxonomic group logit.Area is the group's log odds transformed relative abundance DoY is the day of the year elapsed.days is the time since initiation of the study. Example data for one parameter is provided below. This has identified a number of parameters that show seasonal variability and another set of parameters that show long term inter-annual variability. I would like to identify whether the long term variation for this variable is predictable from the multivariate El-Nino Southern oscillation index (MEI), also provided below. Unfortunately I am not sure how I would test this. Any suggestions? Sincerely, Jacob RDate,elapsed.days,DoY,MEI,logit.Area 2000-08-17,0,230,-0.146,-0.0504416284090222 2000-08-17,0,230,-0.146,-0.0504416284090222 2000-08-17,0,230,-0.146,-0.0504416284090222 2000-08-17,0,230,-0.146,-0.0504416284090222 2000-09-18,32,262,-0.249,-0.467074759836293 2000-09-18,32,262,-0.249,-0.467074759836293 2000-09-18,32,262,-0.249,-0.467074759836293 2000-10-09,53,283,-0.382,-0.348402739936071 2000-10-09,53,283,-0.382,-0.348402739936071 2000-10-09,53,283,-0.382,-0.348402739936071 2000-10-09,53,283,-0.382,-0.348402739936071 2000-12-08,113,343,-0.581,-0.15334106437543 2000-12-08,113,343,-0.581,-0.15334106437543 2000-12-08,113,343,-0.581,-0.15334106437543 2000-12-08,113,343,-0.581,-0.15334106437543 2001-01-16,152,16,-0.54,-0.673220475271774 2001-01-16,152,16,-0.54,-0.673220475271774 2001-01-16,152,16,-0.54,-0.673220475271774 2001-01-16,152,16,-0.54,-0.673220475271774 2001-02-05,172,36,-0.699,-0.61864178154517 2001-02-05,172,36,-0.699,-0.61864178154517 2001-02-05,172,36,-0.699,-0.61864178154517 2001-02-05,172,36,-0.699,-0.61864178154517 2001-03-19,214,78,-0.591,0.0323819413186415 2001-03-19,214,78,-0.591,0.0323819413186415 2001-03-19,214,78,-0.591,0.0323819413186415 2001-03-19,214,78,-0.591,0.0323819413186415 2001-04-02,228,92,-0.149,-0.412683387668689 2001-04-02,228,92,-0.149,-0.412683387668689 2001-04-02,228,92,-0.149,-0.412683387668689 2001-04-02,228,92,-0.149,-0.412683387668689 2001-05-09,265,129,0.151,-0.252047467102658 2001-05-09,265,129,0.151,-0.252047467102658 2001-05-09,265,129,0.151,-0.252047467102658 2001-05-09,265,129,0.151,-0.252047467102658 2001-06-23,310,174,-0.082,-0.278011323004986 2001-06-23,310,174,-0.082,-0.278011323004986 2001-06-23,310,174,-0.082,-0.278011323004986 2001-06-23,310,174,-0.082,-0.278011323004986 2001-07-27,344,208,0.223,-0.965949833146247 2001-07-27,344,208,0.223,-0.965949833146247 2001-07-27,344,208,0.223,-0.965949833146247 2001-07-27,344,208,0.223,-0.965949833146247 2001-08-24,372,236,0.357,-0.395916644785599 2001-08-24,372,236,0.357,-0.395916644785599 2001-08-24,372,236,0.357,-0.395916644785599 2001-08-24,372,236,0.357,-0.395916644785599 2001-09-28,407,271,-0.131,-0.568070396068749 2001-09-28,407,271,-0.131,-0.568070396068749 2001-09-28,407,271,-0.131,-0.568070396068749 2001-09-28,407,271,-0.131,-0.568070396068749 2001-10-29,438,302,-0.276,0.00433931080895083 2001-10-29,438,302,-0.276,0.00433931080895083 2001-10-29,438,302,-0.276,0.00433931080895083 2001-10-29,438,302,-0.276,0.00433931080895083 2001-11-19,459,323,-0.181,-0.596502420009713 2001-11-19,459,323,-0.181,-0.596502420009713 2001-11-19,459,323,-0.181,-0.596502420009713 2001-11-19,459,323,-0.181,-0.596502420009713 2001-12-17,487,351,0.003,-0.806445429137077 2001-12-17,487,351,0.003,-0.806445429137077 2001-12-17,487,351,0.003,-0.806445429137077 2001-12-17,487,351,0.003,-0.806445429137077 2002-01-14,515,14,-0.05,-1.04911451755897 2002-01-14,515,14,-0.05,-1.04911451755897 2002-01-14,515,14,-0.05,-1.04911451755897 2002-01-14,515,14,-0.05,-1.04911451755897 2002-02-20,552,51,-0.206,-0.00546905363182486 2002-02-20,552,51,-0.206,-0.00546905363182486 2002-02-20,552,51,-0.206,-0.00546905363182486 2002-02-20,552,51,-0.206,-0.00546905363182486 2002-03-15,575,74,-0.187,-0.543418349329509 2002-03-15,575,74,-0.187,-0.543418349329509 2002-03-15,575,74,-0.187,-0.543418349329509 2002-04-22,613,112,0.337,-0.200012640323267 2002-04-22,613,112,0.337,-0.200012640323267 2002-04-22,613,112,0.337,-0.200012640323267 2002-04-22,613,112,0.337,-0.200012640323267 2002-05-20,641,140,0.766,0.0464233195400769 2002-05-20,641,140,0.766,0.0464233195400769 2002-05-20,641,140,0.766,0.0464233195400769 2002-05-20,641,140,0.766,0.0464233195400769 2002-06-19,671,170,0.854,-0.647290280298141 2002-06-19,671,170,0.854,-0.647290280298141 2002-06-19,671,170,0.854,-0.647290280298141 2002-06-19,671,170,0.854,-0.647290280298141 2002-07-11,693,192,0.576,-0.422237525936998 2002-07-11,693,192,0.576,-0.422237525936998
Re: [R-sig-eco] Cosinor with data that trend over time
Thanks Gavin, Your suggestion seems promising. I don't think I'll do the derivatives analysis at this time so no hurry on those codes. New question. I am wondering if there are additional considerations when making multiple comparasons with a gamm model. I have been testing a large number of parameters to see which of say 20 environmental variables are seasonal. I would expect a number of these variables to be seasonal but based on this conversation http://stats.stackexchange.com/questions/49052/are-splines-overfitting-the-data I am nervous about the variety of possible spline fits somehow compounding the problem of testing multiple variables for seasonality. Any thoughts? -Jacob On Wed, Mar 26, 2014 at 2:34 PM, Gavin Simpson ucfa...@gmail.com wrote: Sorry about the errors (typos, not syntax errors) - I was forgetting that you'd need to use `gamm()` and hence access the `$gam` component I don't follow the point about a factor trending up or down. You shouldn't try to use the `$lme` part of the model for this. `summary(mod$gam)` should be sufficient, but as it relates to a spline, this is more a test of whether the spline is different from a horizontal, flat, null line. The problem with splines is that the trend need not just be trending up or down. In the past, to convey where change in the trend occurs I have used the first derivative of the fitted spline and looked for where in time the 95% confidence interval on the first derivative of the spline doesn't include zero; that shows the regions in time where the trend is significantly increasing or decreasing. I cover how to do this in a blog post I wrote: http://www.fromthebottomoftheheap.net/2011/06/12/additive-modelling-and-the-hadcrut3v-global-mean-temperature-series/ the post contains links to the R code used for the derivatives etc, though it is a little more complex in the case of a model with a trend spline and seasonal spline. I'm supposed to have updated those codes and the post because several people have asked me how I do the analysis for models with multiple spline terms. If you can't get the code to work for your models, ping me back and I'll try to move that to the top of my TO DO list. Note the `Xs(time)Fx1` entries in the `summary(mod$lme)` table refer to the basis functions that represent the spline or at least to some part of those basis functions. You can't really make much practical use out of those values are they relate specifically to way the penalised regression spline model has been converted into an equivalent linear mixed effect form. HTH G On 26 March 2014 12:10, Jacob Cram cramj...@gmail.com wrote: Thanks again Gavin, this works. gamm() also models the long term trend with a spline s(Time), which is great. I would still like though, to be able to say whether the factor is trending up or down over time. Would it be fair to query summary(mod$lme)$tTable and to look at the p-value and Value corresponding Xs(time)Fx1 value to identify such a trend? Also, here are a few syntax corrections on the code provided in the last email: 1)visual appoach plot(mod$gam, pages = 1) 2) quantitative approach pred - predict(mod$gam, newdata = newdat, type = terms) take - which(pred[,s(DoY)] == max(pred[,s(DoY)])) or take - as.numeric(which.max(pred[,s(DoY)])) Cheers, -Jacob On Wed, Mar 26, 2014 at 9:46 AM, Gavin Simpson ucfa...@gmail.com wrote: 1) Visually - unless it actually matters exactly on which day in the year the peak is observed? If visually is OK, just do `plot(mod, pages = 1)` to see the fitted splines on a single page. See `?plot.gam` for more details on the plot method. 2) You could generate some new data to predict upon as follows: newdat - data.frame(DoY = seq_len(366), time = mean(foo$time)) Then predict for those new data but collect the individual contributions of the spline terms to the predicted value rather than just the final prediction pred - predict(mod, newdata = newdat, type = terms) Then find the maximal value of the DoY contribution take - which(pred$DoY == max(pred$DoY)) newdat[take, , drop = FALSE] You could use take - which.max(pred$DoY) instead of the `which()` I used, but only if there is a single maximal value. This works because the spline terms in the additive model are just that; additive. Hence because you haven't let the DoY and time splines interact (in the simple model I mentioned, it is more complex if you allow these to interact as you then need to predict DoY for each years worth of time points), you can separate DoY from the other terms. None of the above code has been tested, but was written off top of my head, but should work or at least get you pretty close to something that works. HTH G On 26 March 2014 10:02, Jacob Cram cramj...@gmail.com wrote: Thanks Gavin, This seems like a promising
Re: [R-sig-eco] Cosinor with data that trend over time
Thanks again Gavin, this works. gamm() also models the long term trend with a spline s(Time), which is great. I would still like though, to be able to say whether the factor is trending up or down over time. Would it be fair to query summary(mod$lme)$tTable and to look at the p-value and Value corresponding Xs(time)Fx1 value to identify such a trend? Also, here are a few syntax corrections on the code provided in the last email: 1)visual appoach plot(mod$gam, pages = 1) 2) quantitative approach pred - predict(mod$gam, newdata = newdat, type = terms) take - which(pred[,s(DoY)] == max(pred[,s(DoY)])) or take - as.numeric(which.max(pred[,s(DoY)])) Cheers, -Jacob On Wed, Mar 26, 2014 at 9:46 AM, Gavin Simpson ucfa...@gmail.com wrote: 1) Visually - unless it actually matters exactly on which day in the year the peak is observed? If visually is OK, just do `plot(mod, pages = 1)` to see the fitted splines on a single page. See `?plot.gam` for more details on the plot method. 2) You could generate some new data to predict upon as follows: newdat - data.frame(DoY = seq_len(366), time = mean(foo$time)) Then predict for those new data but collect the individual contributions of the spline terms to the predicted value rather than just the final prediction pred - predict(mod, newdata = newdat, type = terms) Then find the maximal value of the DoY contribution take - which(pred$DoY == max(pred$DoY)) newdat[take, , drop = FALSE] You could use take - which.max(pred$DoY) instead of the `which()` I used, but only if there is a single maximal value. This works because the spline terms in the additive model are just that; additive. Hence because you haven't let the DoY and time splines interact (in the simple model I mentioned, it is more complex if you allow these to interact as you then need to predict DoY for each years worth of time points), you can separate DoY from the other terms. None of the above code has been tested, but was written off top of my head, but should work or at least get you pretty close to something that works. HTH G On 26 March 2014 10:02, Jacob Cram cramj...@gmail.com wrote: Thanks Gavin, This seems like a promising approach and a first pass suggests it works with this data. I can't quite figure out how I would go about interrogating the fitted spline to deterine when the peak value happens with respect to DoY. Any suggestions? -Jacob On Tue, Mar 25, 2014 at 9:06 PM, Gavin Simpson ucfa...@gmail.com wrote: I would probably attack this using a GAM modified to model the residuals as a stochastic time series process. For example require(mgcv) mod - gamm(y ~ s(DoY, bs = cc) + s(time), data = foo, correlation = corCAR1(form = ~ time)) where `foo` is your data frame, `DoY` is a variable in the data frame computed as `as.numeric(strftime(RDate, format = %j))` and `time` is a variable for the passage of time - you could do `as.numeric(RDate)` but the number of days is probably large as we might encounter more problems fitting the model. Instead you might do `as.numeric(RDate) / 1000` say to produce values on a more manageable scale. The `bs = cc` bit specifies a cyclic spline applicable to data measured throughout a year. You may want to fix the start and end knots to be days 1 and days 366 respectively, say via `knots = list(DoY = c(0,366))` as an argument to `gam()` [I think I have this right, specifying the boundary knots, but let me know if you get an error about the number of knots]. The residuals are said to follow a continuois time AR(1), the irregular-spaced counter part to the AR(1), plus random noise. There may be identifiability issues as the `s(time)` and `corCAR1()` compete to explain the fine-scale variation. If you hit such a case, you can make an educated guess as to the wiggliness (degrees of freedom) for the smooth terms based on a plot of the data and fix the splines at those values via argument `k = x` and `fx = TRUE`, where `x` in `k = x` is some integer value. Both these go in as arguments to the `s()` functions. If the trend is not very non-linear you can use a low value 1-3 here for x and for the DoY term say 3-4 might be applicable. There are other ways to approach this problem of identifiability, but that would require more time/space here, which I can go into via a follow-up if needed. You can interrogate the fitted splines to see when the peak value of the `DoY` term is in the year. You can also allow the seasonal signal to vary in time with the trend by allowing the splines to interact in a 2d-tensor product spline. Using `te(DoY, time, bs = c(cc,cr))` instead of the two `s()` terms (or using `ti()` terms for the two marginal splines and the 2-d spline). Again you can add in the `k` = c(x,y), fx = TRUE)` to the `te()` term where `x` and `y` are the dfs for each
[R-sig-eco] Cosinor with data that trend over time
Hello all, I am thinking about applying season::cosinor() analysis to some irregularely spaced time series data. The data are unevenly spaced, so usual time series methods, as well as the nscosinor() function are out. My data do however trend over time and I am wondering if I can feed date as a variable into my cosinor analyis. In the example below, then I'd conclude then that the abundances are seasonal, with maximal abundance in mid June and furthermore, they are generally decreasing over time. Can I use both time variables together like this? If not, is there some better approach I should take? Thanks in advance, -Jacob For context lAbundance is logg-odds transformed abundance data of a microbial species in a given location over time. RDate is the date the sample was collected in the r date format. res - cosinor(lAbundance ~ RDate, date = RDate, data = lldata) summary(res) Cosinor test Number of observations = 62 Amplitude = 0.58 Phase: Month = June , day = 14 Low point: Month = December , day = 14 Significant seasonality based on adjusted significance level of 0.025 = TRUE summary(res$glm) Call: glm(formula = f, family = family, data = data, offset = offset) Deviance Residuals: Min 1Q Median 3Q Max -2.3476 -0.6463 0.1519 0.6574 1.9618 Coefficients: Estimate Std. Error t value Pr(|t|) (Intercept) 0.0115693 1.8963070 0.006 0.99515 RDate -0.0003203 0.0001393 -2.299 0.02514 * cosw-0.5516458 0.1837344 -3.002 0.00395 ** sinw 0.1762904 0.1700670 1.037 0.30423 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1 (Dispersion parameter for gaussian family taken to be 0.9458339) Null deviance: 70.759 on 61 degrees of freedom Residual deviance: 54.858 on 58 degrees of freedom AIC: 178.36 Number of Fisher Scoring iterations: 2 llldata as a csv below RDate,lAbundance 2003-03-12,-3.3330699059335 2003-05-21,-3.04104625745886 2003-06-17,-3.04734680029566 2003-07-02,-4.18791034708572 2003-09-18,-3.04419201802053 2003-10-22,-3.13805060873929 2004-02-19,-3.80688269144794 2004-03-17,-4.50755507726145 2004-04-22,-4.38846502542992 2004-05-19,-3.06618649442674 2004-06-17,-5.20518774876304 2004-07-14,-3.75041853151097 2004-08-25,-3.67882486716196 2004-09-22,-5.22205827512234 2004-10-14,-3.99297508670535 2004-11-17,-4.68793287601157 2004-12-15,-4.31712380781011 2005-02-16,-4.30893550479904 2005-03-16,-4.05781773988454 2005-05-11,-3.94746237402035 2005-07-19,-4.91195185391358 2005-08-17,-4.93590576323119 2005-09-15,-4.85820800095518 2005-10-20,-5.22956391101343 2005-12-13,-5.12244047315448 2006-01-18,-3.04854660925046 2006-02-22,-6.77145858348375 2006-03-29,-4.33151493849021 2006-04-19,-3.36152357710535 2006-06-20,-3.09071584142593 2006-07-25,-3.31430484483825 2006-08-24,-3.09974933041469 2006-09-13,-3.33288992218458 2007-12-17,-4.19942661980677 2008-03-19,-3.86146499633625 2008-04-22,-3.36161599919095 2008-05-14,-4.30878307213324 2008-06-18,-3.74372448768828 2008-07-09,-4.65951429661651 2008-08-20,-5.35984647704619 2008-09-22,-4.78481898261137 2008-10-20,-3.58588161980965 2008-11-20,-3.10625125552057 2009-02-18,-6.90675477864855 2009-03-11,-3.43446932013368 2009-04-23,-3.82688066341466 2009-05-13,-4.44885332005661 2009-06-18,-3.97671552612412 2009-07-09,-3.40185954003936 2009-08-19,-3.44958231694091 2009-09-24,-3.86508161094726 2010-01-28,-4.95281587967569 2010-02-11,-3.78064756876257 2010-03-24,-3.5823501176064 2010-04-27,-4.3363571587 2010-05-17,-3.90545735473055 2010-07-21,-3.3147176517321 2010-08-11,-4.53218360860017 2010-10-21,-6.90675477864855 2010-11-23,-6.90675477864855 2010-12-16,-6.75158176094352 2011-01-11,-6.90675477864855 [[alternative HTML version deleted]] ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[R-sig-eco] Mantel test with multiple imputation, P values
Dear List,
I have an environmental data set with several missing values that I
am trying to relate to a community structure data set using a mantel
test. One solution to the missing data problem seems to be multiple
imputation; I am using the Amelia package. This generates several (five in
this example) imputed data sets. I can run mantel on each of these and
come up with five similar but not identical solutions. I figure I can
average the mantel rho values. However, I am not sure what to do about the
P values. From looking around online, it looks like I shouldn't take the
average of p values. I found this reference
http://missingdata.lshtm.ac.uk/index.php?option=com_contentview=articleid=164:combining-p-values-from-multiple-imputationscatid=57:multiple-imputationItemid=98
that seems to have promising suggestions, but I can't seem to figure out
how I'd implement any of these in R. I was hoping somebody might have a
suggestion for how I could combine my p values. One option, I think would
be to take the highest (worst) p value (in the example below, this would be
p = 0.012). However for large numbers of imputations, I am believe that
this method might be to conservative. Another option might be to take the
p value corresponding to the median rho score (in the example below this
would be p =0.008). Thoughts?
-Jacob
##Example Code Below
require(Amelia)
require(vegan)
require(ecodist)
##Species data matrix with environmental data that are missing some values.
data(varespec)
data(varechem)
varechem.missings - varechem[,c(N, P, K)]
varechem.missings[c(1,5, 7, 15, 20),1] - NA
varechem.missings[c(1,2, 9, 21), 2] - NA
#I multiply impute the missing values with the Amelia package
imps - 5
#imps - 25
varechem.amelia - amelia(varechem.missings, m = imps)
#for each imputation of the environmental data I run a mantel test and save
#the results to mresults
mresults - NULL
for(i in 1:imps){
varespec.dist - vegdist(varespec)
varechem.am.dist - dist(varechem.amelia$imputations[[i]])
mresults - rbind(mresults,
(ecodist::mantel(varespec.dist~varechem.am.dist)))
}
mresults
##mantelr pval1 pval2 pval3 llim.2.5% ulim.97.5%
## [1,] 0.2137656 0.008 0.993 0.008 0.1015176 0.3389979
## [2,] 0.2162388 0.011 0.990 0.011 0.1207528 0.3346554
## [3,] 0.2149556 0.012 0.989 0.016 0.1319943 0.3279028
## [4,] 0.2101820 0.009 0.992 0.012 0.1217293 0.3288272
## [5,] 0.2135279 0.006 0.995 0.006 0.1130386 0.3359864
#based on these results what would be a reasonable p value to report for
the environmental parameters relating to the community structure?
##end example
[[alternative HTML version deleted]]
___
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
Re: [R-sig-eco] Mantel test with multiple imputation, P values
Krzystof,
Thank you for your reply. It is promising that you think that Li's
method should work for imputed mantel results. That said, I am a bit over
my head with the math in the Li et al reference. Could you (or anyone on
this list) provide an R-code example of how D would be calculated from the
output of several mantel tests?
Also, and forgive my ignorance if this statement is coming from the wrong
direction, It is my understanding that mantel P values are generally
calculated by actually permuting the rows of the similarity matrix and
re-running the mantel test a given number of times (usually 1000).
Accordingly, as far as I can tell there is no explicitly generated F value
corresponding to a mantel p value. It seems like Li's method assumes I am
generating P from an F table. Would it be appropriate to back calculate F
from P, k and m?
Thanks again,
Jacob
On Sun, Aug 18, 2013 at 2:19 PM, Krzysztof Sakrejda
krzysztof.sakre...@gmail.com wrote:
Hi Jacob, comments below.
On Aug 18, 2013 2:31 PM, Jacob Cram cramj...@gmail.com wrote:
Dear List,
I have an environmental data set with several missing values that I
am trying to relate to a community structure data set using a mantel
test. One solution to the missing data problem seems to be multiple
imputation; I am using the Amelia package. This generates several (five
in
this example) imputed data sets. I can run mantel on each of these and
come up with five similar but not identical solutions. I figure I can
average the mantel rho values. However, I am not sure what to do about
the
P values. From looking around online, it looks like I shouldn't take the
average of p values. I found this reference
http://missingdata.lshtm.ac.uk/index.php?option=com_contentview=articleid=164:combining-p-values-from-multiple-imputationscatid=57:multiple-imputationItemid=98
that seems to have promising suggestions, but I can't seem to figure out
how I'd implement any of these in R.
So following that link and reading the Li et al. reference it looks as
though the procedure is well described at the top of page 71. You get your
parameter estimate from the usual procedure. The test statistic, written as
D, is the distance between the null value and the estimated value with
some scaling specified in eq. 1.17. They use the F distribution and k and m
(the number of imputations) degrees of freedom. I don't think you need to
reinvent some inferior ecologists-only procedure for this.
Krzysztof
I was hoping somebody might have a
suggestion for how I could combine my p values. One option, I think
would
be to take the highest (worst) p value (in the example below, this would
be
p = 0.012). However for large numbers of imputations, I am believe that
this method might be to conservative. Another option might be to take
the
p value corresponding to the median rho score (in the example below this
would be p =0.008). Thoughts?
-Jacob
##Example Code Below
require(Amelia)
require(vegan)
require(ecodist)
##Species data matrix with environmental data that are missing some
values.
data(varespec)
data(varechem)
varechem.missings - varechem[,c(N, P, K)]
varechem.missings[c(1,5, 7, 15, 20),1] - NA
varechem.missings[c(1,2, 9, 21), 2] - NA
#I multiply impute the missing values with the Amelia package
imps - 5
#imps - 25
varechem.amelia - amelia(varechem.missings, m = imps)
#for each imputation of the environmental data I run a mantel test and
save
#the results to mresults
mresults - NULL
for(i in 1:imps){
varespec.dist - vegdist(varespec)
varechem.am.dist - dist(varechem.amelia$imputations[[i]])
mresults - rbind(mresults,
(ecodist::mantel(varespec.dist~varechem.am.dist)))
}
mresults
##mantelr pval1 pval2 pval3 llim.2.5% ulim.97.5%
## [1,] 0.2137656 0.008 0.993 0.008 0.1015176 0.3389979
## [2,] 0.2162388 0.011 0.990 0.011 0.1207528 0.3346554
## [3,] 0.2149556 0.012 0.989 0.016 0.1319943 0.3279028
## [4,] 0.2101820 0.009 0.992 0.012 0.1217293 0.3288272
## [5,] 0.2135279 0.006 0.995 0.006 0.1130386 0.3359864
#based on these results what would be a reasonable p value to report for
the environmental parameters relating to the community structure?
##end example
[[alternative HTML version deleted]]
___
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[[alternative HTML version deleted]]
___
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[R-sig-eco] cocorresp to determine best of several predictor communities
Dear List Members, I am using the 'cocorresp' package and trying to compare several pairs of communities. My goal would be to see which of several communities best predicts a final community. Say, for instance, that I have four communities A, B, C and D. I want to know which community best predicts D. To further complicate things, I have more data points for some communities than others. For instance communities A and B and D have 100 sites associated with them, while C only was measured at 75 sites (lets imagine it wasn't measured in the last 25 sites). So I run A.pred - coca(x =A, y = D, method = pred, reg.method = eigen n.axes = 10) B.pred - coca(x =B, y = D, method = pred, reg.method = eigen n.axes = 10) C.pred - coca(x =C, y = D[1:75,], method = pred, reg.method = eigen n.axes = 10) I then apply crossvalidation, as in (Braak Schaffers 2004). A.crossval - crossval(x = A.pred, n.axes = 10) A.cvax - which.max(A.crossval$CVfit) A.cvmax - max(A.crossval$CVfit) B.crossval - crossval(x = B.pred, n.axes = 10) B.cvax - which.max(B.crossval$CVfit) B.cvmax - max(B.crossval$CVfit) C.crossval - crossval(x = C.pred, n.axes = 10) C.cvax - which.max(C.crossval$CVfit) C.cvmax - max(C.crossval$CVfit) Thus X.cvax is the number of axes that give the best crossvalidation score for community X predicting community D. X.cvax max is that best crossvalidation score. I think that if X.cvmax is positive, that suggests that the model can predict come fraction of the structure of community D from community X (where X is the community (A, B or C) in question) (as suggested in Schaffers et al 2008). Now, I want to know whether A, B or C best predicts D. Because C is only 75 data points, I expect A.cvmax and B.cvmax should be inflated relative to C.cvmax, so I shouldn't be able to use these values to cross compare predictive power. One approach that I think might work is to run permutest to generate percent fits for each axis that crossvalidation suggests I should include in the model. A.perm - permutest(mo.pred, n.axes = A.cvax) B.perm - permutest(mo.pred, n.axes = B.cvax) C.perm - permutest(mo.pred, n.axes = C.cvax) I can then ask what percentage of the community of X predicts what percent of community Y using the pcent.fit value from the permutest objects. I think this value is independent of sample size, but am not sure. Does anybody else know? I could either do this for the first axis, or for the sum all axes in the model. I'm not sure if the second option is legitimate Option 1 A.ax1 - A.perm$pcent.fit[1] B.ax1 - B.perm$pcent.fit[1] C.ax1 - C.perm$pcent.fit[1] Option 2 A.all - sum(A.perm$pcent.fit[1:A.cvax]) B.all - sum(A.perm$pcent.fit[1:B.cvax]) C.all - sum(B.perm$pcent.fit[1:C.cvax]) So my first question is, when comparing communities with different but overlapping sample sizes(all 75 sample sites in C are the same sample sites as the first 75 in A and B), if A.all B.all C.all or if A.ax1 B.ax1 C.ax1, can I then say that A is the best predictor community for community D? More specifically, is this approach a legitimate way of comparing communities with different sample sizes in which the sample locations overlap. If not is there a better way of making such a comparison? Perhaps the co-correspondence method is not appropriate for making this kind of comparison. I notice that most literature to date uses co-correspondence in conjunction with cca to determine whether a different community or environmental parameters best predict a given community. If co-correspondence is not appropriate, could somebody explain why and suggest a better method? References: Braak CJF ter, Schaffers AP. (2004). Co-Correspondence Analysis: A New Ordination Method to Relate Two Community Compositions. Ecology 85:834846. Schaffers AP, Raemakers IP, Sýkora KV, Braak CJF ter. (2008). Arthropod Assemblages Are Best Predicted by Plant Species Composition. Ecology 89:782794. Thanks everyone for your consideration in this matter. Sincerely, Jacob Cram Graduate Student Department of Biological Sciences University of Southern California [[alternative HTML version deleted]] ___ R-sig-ecology mailing list R-sig-ecology@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
